Patent Grant
12175568
The present invention is generally related to data manipulation and more specifically the visualization and manipulation of data.
In computing, a graph is an abstract data structure including nodes and edges. A graph contains a set of nodes connected by one or more edges. Values can be associated with the nodes and/or the edges. A graph data structure is an implementation of the mathematical concept of a graph, which is a representation of a set of objects where some pairs of the objects are connected by links. Graphs can be undirected, where an edge indicates a relationship between two nodes within the graph. Graphs can also be directed, where an edge indicates a relationship between a first node and a second node within the graph, but not the corresponding relationship between the second node and the first node.
Systems and methods for visualizing and manipulating graph databases in accordance embodiments of the invention are disclosed. In one embodiment of the invention, a graph database manipulation device includes a processor and a memory configured to store a graph database manipulation application, wherein the graph database manipulation application configures the processor to obtain a graph database, wherein the graph database includes a set of nodes and a set of edges, wherein an edge in a set of edges defines a relationship between a first node in the set of nodes and a second node in the set of nodes and an edge includes edge weight metadata and edge display metadata, wherein the edge display metadata describes the spatial relationship between the first node and the second node, determine a source node within the set of nodes, locate a set of related nodes based on the source node and the set of edges, where a related node in the set of related nodes has an edge in the set of edges indicating a relationship between the related node and the source node, recursively locate a set of sub-related nodes based on the set of related nodes and the set of edges, where a sub-related node in the set of sub-related nodes has an edge in the set of edges indicating a relationship between a related node in the set of related nodes and the sub-related node, generate a representation of the set of related nodes from the perspective of the source node, where the representation of a related node in the subset of the set of related nodes is based on the edge weight metadata and the edge display metadata from the edge defining the relationship between the particular related node and the source node, and recursively update the generated representation of the set of sub-related nodes from the perspective of the source node and the set of related nodes, where the representation of a sub-related node in the set of sub-related nodes within the generated representation is recursively based on the edge weight metadata and the edge display metadata from the edge defining the relationship between the particular sub-related node and its predecessor nodes.
In an additional embodiment of the invention, the system further includes a display device and configured to display a visualization of a representation of nodes and edges within the graph database, wherein the graph database manipulation application further configures the processor to display the generated representation using the display device.
In another embodiment of the invention, the display of the generated representation further includes performing a recursive shift based on the relationship between the related nodes in the set of related nodes and the edge display metadata for the subset of edges defining the relationship between pairs of the related nodes in the set of related nodes.
In yet another additional embodiment of the invention, the display of the generated representation further includes performing a recursive transformation based on the relationship between the related nodes in the set of related nodes and the edge display metadata for the subset of edges defining the relationship between pairs of the related nodes in the set of related nodes.
In still another additional embodiment of the invention, the system further includes an input device configured to receive graph manipulation data, wherein the graph database manipulation application further configures the processor to modify the nodes and edges within the graph database based on the graph manipulation data and refresh the generated representation of the source node and the set of related nodes based on the modified graph database.
In yet still another additional embodiment of the invention, a node includes permission metadata, where the permission metadata describes a set of nodes that have access to the node and the graph database manipulation application further configures the processor to locate the set of related nodes for the source based on the permission metadata for the nodes in the set of related nodes.
In yet another embodiment of the invention, the graph database manipulation application further configures the processor to recursively locate the sub-related nodes in the set of sub-related nodes based on the permission data for the sub-related nodes.
In still another embodiment of the invention, the recursive location of sub-related nodes from a related node further includes receiving a set of related edges from a node having an edge in common with the related node based on the permission metadata for the node.
In yet still another embodiment of the invention, the edge weight metadata is a complex number having a real component and an imaginary component.
In yet another additional embodiment of the invention, the edge weight metadata represents a property selected from the group consisting of a spatial position, a color, and a size.
In still another additional embodiment of the invention, the meaning of the edge weight metadata is based on the generated representation of the nodes associated with the edge including the edge weight metadata.
In yet still another additional embodiment of the invention, the recursive update of the visualized representation is based on an accumulation of the weights of the sub-related nodes.
In yet another embodiment of the invention, a portion of the edge display metadata is convertible to a binary string.
In still another embodiment of the invention, the edge display metadata describes the relative layout of the nodes associated with the edge including the edge display metadata and wherein the generation of the representation of the set of related nodes and the set of sub-related nodes based on the perspective of the source node further includes recursively calculating the position of the representation of a sub-related node based on the edge display metadata for the sub-related node and the edge display metadata for nodes within the set of related nodes that are predecessor nodes to the sub-related node.
In yet still another embodiment of the invention, at least one third-party node in the set of nodes represents a third-party data source device and the at least one third-party node includes node metadata retrieved from the third-party data source device.
In yet another additional embodiment of the invention, the edge weight metadata for the edges in the set of edges that are connected to the at least one third-party node is based on the latency associated with retrieving the node metadata from the third-party data source device.
In still another additional embodiment of the invention, the edge weight metadata for the edges in the set of edges that are connected to the at least one third-party node is based on the latency associated with retrieving edge metadata from the third-party data source device.
In yet still another additional embodiment of the invention, the edge display metadata is calculated based on the related nodes and the sub-related nodes.
In yet another embodiment of the invention, the display metadata includes time data describing a time associated with the edge.
In still another embodiment of the invention, the generated representation includes a historical narrative of the set of nodes based on the time data.
In yet still another embodiment of the invention, at least one generated representation includes a partially overlapping subset of at least one other generated representation.
In yet another additional embodiment of the invention, at least one node in the set of nodes is configured to execute actions based on a received request.
In still another additional embodiment of the invention, the received request includes a request for the set of nodes related to the at least one node configured to execute actions and the at least one node configured to execute actions to identify nodes connected to the at least one node by at least one edge, generate a set of sub-related nodes based on the identified nodes and edges, where the set of sub-related nodes includes the sub-related nodes and the edges associated with the sub-related nodes and the at least one node, and transmit a portion of the generated set of sub-related nodes based on the received request.
Yet another embodiment of the invention includes method for visualizing graph databases including obtaining a graph database using a graph database manipulation device, wherein the graph database includes a set of nodes and a set of edges, wherein an edge in a set of edges defines a relationship between a first node in the set of nodes and a second node in the set of nodes and an edge includes edge weight metadata and edge display metadata, wherein the edge display metadata describes the spatial relationship between the first node and the second node, determining a source node within the set of nodes using the graph database manipulation device, locating a set of related nodes based on the source node and the set of edges using the graph database manipulation device, where a related node in the set of related nodes has an edge in the set of edges indicating a relationship between the related node and the source node, recursively locating a set of sub-related nodes based on the set of related nodes and the set of edges using the graph database manipulation device, where a sub-related node in the set of sub-related nodes has an edge in the set of edges indicating a relationship between a related node in the set of related nodes and the sub-related node, generating a representation of the set of related nodes from the perspective of the source node using the graph database manipulation device, where the representation of a related node in the subset of the set of related nodes is based on the edge weight metadata and the edge display metadata from the edge defining the relationship between the particular related node and the source node, and recursively updating the generated representation of the set of sub-related nodes from the perspective of the source node and the set of related nodes using the graph database manipulation device, where the representation of a sub-related node in the set of sub-related nodes within the generated representation is recursively based on the edge weight metadata and the edge display metadata from the edge defining the relationship between the particular sub-related node and its predecessor nodes.
Still another embodiment of the invention includes a graph database manipulation device including a processor and a memory configured to store a graph database manipulation application, wherein the graph database manipulation application configures the processor to obtain a graph database, wherein the graph database includes a set of nodes and a set of edges, identify a region of interest within a graph described by the graph database, construct a feature space from the region of interest, and extract explanatory variables from the feature space.
In yet another additional embodiment of the invention, constructing a feature space further includes integrating first-order connections, integrating first-order weights, integrating higher-order connections, and integrating higher-order weights.
In still another additional embodiment of the invention, the graph database manipulation application further directs the processor to extract at least one unknown explanatory variable from the feature space.
In yet still another additional embodiment of the invention, extracting the at least one unknown explanatory variable from the feature space includes applying machine learning technique on a subgraph.
In yet another embodiment of the invention, the predictive power of the at least one unknown explanatory variable is determined using a statistical classifier.
In still another embodiment of the invention, the at least one unknown explanatory variable is incorporated into the graph database.
In yet still another embodiment of the invention, the graph database manipulation application further configures the processor to generate at least one supernode.
In yet another additional embodiment of the invention, at least one of the at least one supernode is a superfeature including data describing at least two features.
In still another additional embodiment of the invention, at least one of the at least one supernode is a superobservation including data describing at least two observations.
In yet still another additional embodiment of the invention, the graph database manipulation application further configures the processor to store the at least one supernode.
In yet another embodiment of the invention, the graph database manipulation application further configures the processor to obtain a tabular data structure including at least one row and at least one column and convert the tabular data structure into a graph database.
In still another embodiment of the invention, the graph database manipulation application further configures the processor to generate a directed acyclic graph from the tabular data structure.
In yet still another embodiment of the invention, each of the at least one rows corresponds to a unique primary key.
In yet another additional embodiment of the invention, each of the at least one columns includes a column header, wherein the column header describes a column type.
In yet still another additional embodiment of the invention, at least one value in at least one of the at least one row is defined as unique, the at least one value appears a plurality of times in the tabular data structure, and the at least one value maps onto a unique node in the graph database.
In yet another embodiment of the invention, at least one value in at least one of the at least one column is defined as unique, the at least one value appears a plurality of times in the tabular data structure, and the at least one value maps onto a unique node in the graph database.
In still another embodiment of the invention, the graph database manipulation application further configures the processor to obtain a hierarchical data structure with attributes and convert the hierarchical data structure into a directed acyclic graph with attributes of the hierarchical data structure mapped onto unique nodes in the directed acyclic graph.
Still another embodiment of the invention includes a method including obtaining a graph database using a graph database manipulation device including a processor and a memory connected to the processor, wherein the graph database includes a set of nodes and a set of edges, identifying a region of interest within a graph described by the graph database using the graph database manipulation device, constructing a feature space from the region of interest using the graph database manipulation device, and extracting explanatory variables from the feature space using the graph database manipulation device.
In yet another additional embodiment of the invention, constructing a feature space further includes integrating first-order connections using the graph database manipulation device, integrating first-order weights using the graph database manipulation device, integrating higher-order connections using the graph database manipulation device, and integrating higher-order weights using the graph database manipulation device.
In still another additional embodiment of the invention, the method further includes obtaining a tabular data structure including at least one row and at least one column using the graph database manipulation device and converting the tabular data structure into a graph database using the graph database manipulation device.
In yet still another additional embodiment of the invention, the method further includes generating a directed acyclic graph from the tabular data structure using the graph database manipulation device.
In yet another embodiment of the invention, each of the at least one rows corresponds to a unique primary key.
In still another embodiment of the invention, each of the at least one columns includes a column header, wherein the column header describes a column type.
Turning now to the drawings, systems and methods for visualizing and manipulating graph databases in accordance with embodiments of the invention are illustrated. Text-based search has enjoyed great success in the last decade, yet paradoxically, it has failed to revolutionize data driven fields such as medicine. It can be difficult for users to conjure a search term if the user does not know exactly what information the user is searching for. Graph databases provide a user an opportunity to explore the relationships within a space of ideas. In this way, graph databases provide users with the ability to locate information that answers the user's question without having to formulate the search query to express the question. Additionally, a user can encode new information into the graph by creating associations, or streamline existing information by pruning associations. These manipulations can improve the graph's usefulness as a record for the same user, and a means of sharing information with other users and computational agents that operate within the graph. In this way, graph database manipulation systems can provide a powerful tool for storing, visualizing, manipulating, and analyzing data using a robust set of visual tools that, in many embodiments, do not require the use of complex code or other queries. Many graph database manipulation systems also include the ability to seamlessly import and export data from existing database systems, including the ability to transform existing queries to and from visual commands that can be used to interact with the graph database. In this way, graph database manipulation systems improve on existing text-based search and relational database systems by improving the ability of the computer itself to analyze and present data, by default and in response to visual commands provided to the graph database manipulation system.
In a variety of embodiments, the analysis of natural data involve extensive use of set operations, such as but not limited to concatenating data sets thought to be similar, excluding outlier sets, and taking differences of sets to find exclusive elements. Graph databases in accordance with embodiments of the invention facilitate the execution of these set operations as successors of a node can be the set of its successors. Accordingly, graph database manipulation systems allow the successors of parent nodes to be combined, differenced, or otherwise manipulated in order to capture the set operations done in the analysis of naturalistic data. In a number of embodiments, these operations can be performed using visualizations of the data, rather than by querying the graph database using text-based queries. In several embodiments, the set of successors of a given node and their own links to each other encode an entire graph (e.g., a subgraph within the graph database) that has the complete set of attributes (e.g. the nodes, edges, and/or metadata) as any other subgraph within the database as well as the entire database itself. Accordingly, any operation designed to run on a given subgraph can also run on any other subgraph, thereby providing a mechanism by which general-purpose operators may be created and extended across the entire graph database. In this sense, a node perspective can extend to not only the immediate sub-nodes, but also recursively to their cross-links. In this way, the node perspective can operate as a fundamental unit of computation within the framework provided by a graph database manipulation system.
Graph database manipulation systems in accordance with embodiments of the invention are configured to visualize and/or manipulate graph databases. Graph databases contain a set of nodes defining concepts and a set of edges indicating relationships between pairs of nodes. The data stored within the nodes and/or edges can be locally stored within the graph database and/or stored on external devices. Graph database manipulation devices are configured to generate visualizations (or other representation) of the graph database from the perspective of one or more nodes (e.g. source node(s)) within the graph database. The graph database can also be visualized from a point outside of the graph. In several embodiments, the visualization becomes the perspective from a node in a new graph that larger by one node. In a variety of embodiments, a point outside of the graph corresponds to the perspective of a graph database manipulation device with respect to the graph database. The generated representation can be any of a variety of visualizations, such as but not limited to a tile-based histogram, a point cloud, an image, and a node-edge diagram, that can be explored and manipulated as appropriate to the requirements of specific applications in accordance with embodiments of the invention. Once a source node has been selected, related nodes are recursively located within the graph database based on the edges connecting the nodes. The number of related nodes so viewed can be limited by a preset threshold, determined dynamically by the resolution or readability limits of the system, or by processing constraints imposed to maintain the graph database manipulation device simultaneously across a network of portals. In a variety of embodiments, the threshold value is based on one or more nodes and/or edges selected within the graph database. The nodes and/or edges can be selected using a variety of techniques as appropriate to the requirements of specific applications in accordance with embodiments of the invention, such as receiving a selection of nodes and/or edges using an input device configured to receive data indicative of node and/or edge selections within a graph database manipulation device. In several embodiments, the nodes and/or edges selected as related nodes can be based on the statistics (e.g. patterns within the relationships between the nodes and edges from the perspective of the source node) of the surrounding data. The visualization of the generated representation includes a representation of the nodes and the layout and visual appearance of the generated representation can be based on edge weight metadata and edge display metadata contained in the edges connecting the nodes being visualized. In several embodiments, the threshold value can be based on the visualized representation of the nodes, e.g. the threshold can be based on readability metric(s) and/or the amount of visualized space the node consumes as displayed using a graph database manipulation device. In a variety of embodiments, a one-to-one mapping exists between the edge display metadata and some particular perspective of the associated data (e.g. the visualized representation of that data). For example, the color of nodes being assigned dynamically from the output of a calculation (for example, k-means clustering) on the node weights in a particular perspective in a particular fashion. In a number of embodiments, the edge display metadata for any entity within the graph (i.e. an edge or a node) can be statically and/or dynamically generated.
Based on the nodes and the edges connecting the nodes, emergent data describing the relationships between the nodes can be determined that provides additional context and/or description of the nodes and edges. This emergent data can be utilized to provide additional insights into the data and/or additional visualizations of the graph database, such as by changing the format in which the data are viewed, or the affordances given to the user for manipulating the graph. In several embodiments, manipulating the graph database includes creating, deleting, and/or modifying links between nodes within the database. In this way, the links between the nodes can be utilized to encode processes, workflows, and any other data as appropriate to the requirements of specific applications of embodiments of the invention. In a variety of embodiments, a one-to-one mapping between the underlying graph database and the visualized representation of the graph database allows for the modifications to the visualized representation to be incorporated into the underlying graph database. In many embodiments, the emergent data includes aggregations of the relationships between the nodes as described by the edges from the perspective of the source node. In several embodiments, time metadata is associated with nodes and/or edges within the graph database. In a number of embodiments, particularly in those with multiple users, the use of time metadata allows the graph database to be visualized as a historical narrative and/or a communication system. Additionally, the time metadata can be utilized to impose ordering and/or sequencing on the visualized representation of the graph database, such as when chaining multiple operators together in the graph that need to act in sequence. However, it should be noted that any metadata can be associated with the nodes and/or edges that can be used to facilitate the visualization and/or manipulation of graph databases as appropriate to the requirements of specific applications of the invention.
In a variety of embodiments, the data associated with a node within the graph database is calculated based on data present within the node and/or nodes related to the node within the graph database. Node messages can be transmitted between nodes within the database in order to request additional and/or updated data from the nodes. These messages can be utilized to facilitate communication between the nodes and/or between graph database manipulation devices visualizing (related) graph database(s) as appropriate to the requirements of specific applications in accordance with embodiments of the invention. In several embodiments, the nodes are configured to transmit node message responses containing the data requested in the node message. The processing and response to node messages can be influenced by permission metadata associated with the nodes. In this way, nodes within the graph database can be utilized as computation devices that provide requested data in response to a received node message.
During the exploration of the graph database, changes can be made to the generated representation(s) that can be applied to the nodes and edges within the graph database. In many embodiments, the nodes and/or edges have permission metadata describing what data (e.g. the node metadata and/or edges connected to the node) a node and/or edge will provide in the recursive location and/or modification of the nodes and/or edges. For example, a node can be connected to several related nodes (where the node metadata for the node is an aggregation of the node metadata for the related nodes) and the permission metadata for the node will return the node metadata but not the edges linking to the related nodes. Permission metadata can be global, based on the source node, and/or based on the related nodes as appropriate to the requirements of specific applications in accordance with embodiments of the invention. The modification of the graph database can respect (or ignore) any permission metadata associated with the nodes and/or edges as appropriate to the requirements of specific applications in accordance with embodiments of the invention. The permission metadata can be utilized to implement a security model within the graph database so that particular nodes and/or graph database manipulation devices only have access to portions of the data stored within the graph database, while the remaining data remains secured away. The security model limits the unauthorized disclosure of potentially sensitive and/or private data that can be stored within the graph database while keeping the sensitive and/or private data within the graph database for other allowable uses. For example, a “gateway” node can act as a bottleneck based on its security permissions, providing a limited set of data to the other nodes within the graph database while limiting access to potentially sensitive nodes related to the gateway node within the graph database.
In a variety of embodiments, the relationships between the nodes and edges within a graph database can be described using an ontology in that the edges represent a relationship between two nodes within the graph database. Graph database manipulation devices are configured to identify an ontology based on the perspective of a source node within the graph database. That is, the ontology describing the relationships between nodes and edges within the graph database depends on the source node utilized to explore the graph database and/or emergent data determined based on the relationships between the nodes from the perspective of the source node. In this way, the same edges and nodes can indicate different relationships based on the perspective of the node utilized to explore the graph database. In a number of embodiments, the ontology of the graph database can be described locally based on the perspective of a source node in that the ontology of the graph database in determined based on the relationships between the nodes and edges (along with any emergent data) within the local perspective of the source node.
In addition to storing data in a graph database, the data and associated relationships described in the nodes and edges can be utilized to generate data structures, including tabular data structures, containing some or all of the data stored in the graph database. This can be useful in a variety of embodiments of the invention, such as allowing traditional data analysis tools to seamlessly interact with the graph database. Similarly, data stored outside of the graph database can be added to the graph database by creating nodes and edges based on the features of the structure storing the data to be imported. In addition to data stored in databases, annotations can be added to the nodes and/or edges of the graph database, thereby enriching the graph database. The data stored in the database can be aggregated and/or sampled to create a variety of sample graphs and/or summary subgraphs, providing a variety of different views of the data and providing facilities for simplifying many of the complex relationships captured in the graph database. In this way, the graph database can serve as an exchange for the original data, human knowledge, and inferred knowledge.
Although the above is described with respect to a graph having edges that connect a pair of nodes, a variety of data structures, including hypergraphs where edges connect any number of nodes within the graph database, can be utilized as appropriate to the requirements of specific applications in accordance with embodiments of the invention. Similarly, a variety of graph database visualizations not specifically described above can be utilized as appropriate to the requirements of specific applications in accordance with embodiments of the invention. Systems and methods for visualizing and manipulating graph databases in accordance with embodiments of the invention are discussed below.
Graph database manipulation systems in accordance with embodiments of the system are configured to facilitate the creation, manipulation, and interaction with graph databases. In a variety of embodiments, graph database manipulation systems include graph database manipulation devices configured to facilitate this functionality. In several embodiments, graph database manipulation device includes a processor and a memory configured to store a graph database manipulation application, wherein the graph database manipulation application configures the processor to obtain a graph database, wherein the graph database includes a set of nodes and a set of edges, wherein an edge in a set of edges defines a relationship between a first node in the set of nodes and a second node in the set of nodes, and metadata describing a numeric value attributed to an edge or node that can be pre-assigned as a static attribute of the node or edge stored in memory and/or calculated as a function of the connection patterns of nodes and edges to be found within one or more degrees of separation to the node or edge. In a number of embodiments, the graph database manipulation application further configures the processor to determine a source node within the set of nodes, locate a set of related edges connected to the source node, locate the nodes to which those edges in turn connect, and, the above constituting the perspective of the source node within the graph, whereby the perspective, encodes all edges connected to the source node, encodes all additional nodes connected to these edges, can encode recursively, in the above manner, all edges and nodes within one or more degrees of separation from the source node, can extend the above recursive encoding, based on the content of the intervening node or edge metadata, can represent overlapping subsets of node data or metadata which can be encoded in other such perspectives, and continuously update the above described generated representations of the set of sub related nodes, edges and metadata from the source node (i.e. the perspective). In several embodiments, the graph database manipulation application permits the creation, deletion, and modification of nodes within the graph database, edges within the graph database, and metadata within the graph database, whereby the so created and/or deleted and/or modified edges and/or metadata will also augment, prune and/or modify the relevant perspective(s) of the graph database.
In a variety of embodiments, graph database manipulation devices also includes a display device configured to display a visualization of a representation of nodes and edges within the graph database, wherein the graph database manipulation application further configures the processor to display the generated representation using the display device. In many embodiments, the display of the generated representation further includes a recursive operation based on the relationship between the related nodes, edges and metadata, whereby the spatial relationship between a source node to the nodes in its perspective can be represented by a graphical operation, such as a shift, scaling, and/or other mathematical transformation based on intervening edges, nodes and metadata such that these graphical operations can also be continued forward recursively to one or more degrees of separation in the graph. In a number of embodiments, graph database manipulation devices also include input device(s) configured to receive graph manipulation data, wherein the graph database manipulation application further configures the processor to modify the nodes and edges within the graph database based on the graph manipulation data and in particular modify nodes and/or edges and/or metadata within the graph such that actions taken by the user are transformed into the domain of the graph database directly, via an inversion of the transformations, and/or via an inversion of the calculations of the metadata described and refresh the generated representation of the source node and the set of related nodes based on the modified graph database.
In several embodiments, the nodes and edges can be utilized in a computing capacity whereby a node can be imbued with a capacity to automatically generate and/or modify edges in its perspective, thereby augmenting or pruning its perspective and gaining access to and/or relinquishing information encoded in the edges, nodes and metadata in the graph by using a rule or mathematical function based on the edges, nodes and metadata encoded in its current perspective. For example, a node within the graph database can correspond to a client device capable of processing received data and transmitting data describing other computing devices to which the client device can communicate. Similarly, a node can include the capability of executing requests for data (e.g. identifying related nodes within the graph database and/or providing access to secondary nodes connected to the node via one or more edges) and returning responses to those requests based on the received request. In this way, a node can dynamically present some or all of the requested data (e.g. subgraph within the graph database) using the computing capability of the node itself. It should be noted, however, that the requested data can also be generated in response to the received request as appropriate to the requirements of specific applications of embodiments of the invention. In a variety of embodiments, the graph database manipulation application further configures the processor to locate the set of related nodes for the source (the perspective of the source) based on the permission metadata for the nodes in the set of related nodes, including permissions assigned to the nodes, and permissions assigned to the edges. In several embodiments, permission data can be encoded as numerical key and/or bitmask metadata and/or graph structural data, including permission nodes that permit connections between nodes in the set of related nodes within their perspective. In many embodiments, the graph database manipulation application further configures the processor to recursively locate the sub related nodes in the set of sub related nodes based on the permission data for the sub related nodes and/or based on permissions. In a number of embodiments, the metadata is a complex number having a real component and an imaginary component. In several embodiments, the metadata represents a property selected spatial position, a color, and/or a size. In a variety of embodiments, the metadata is a binary string.
In many embodiments, the determination of the extent of the recursive perspective can be based upon the product of the accumulated weights of the magnitude of the edge weight metadata. In a variety of embodiments, the edge display metadata describes the relative layout of the nodes associated with the edge including the edge display metadata and the generation of the representation of the set of related nodes and the set of sub related nodes based on the perspective of the source node further includes recursively calculating the position of the representation of a sub related node based on the edge display metadata for the sub related node and the edge display metadata for nodes within the set of related nodes that are predecessor nodes to the sub related node. In several embodiments, the metadata available in a given perspective can be approximated such that an approximated perspective can be displayed to the user, an approximated version of the edge weight metadata can be used for calculations based on the graph database.
In a number of embodiments, at least one third party node in the set of nodes represents a third party data source device and the at least one third party node includes node metadata retrieved from the third party data source device. In many embodiments, the edge weight metadata for the edges in the set of edges that are connected to the at least one third party node can be based on the time of acquiring data about the node, the latency associated with retrieving the node metadata from the third party data source device, and/or another property related to time and/or space from which other properties related to time and space, such as order or sequence, can be interpreted.
Graph database manipulation systems in accordance with embodiments of the invention are configured to visualize and manipulate graph databases. A conceptual illustration of a graph database manipulation system in accordance with an embodiment of the invention is shown in
The graph database manipulation device 110 is configured to obtain a graph database (or a portion thereof) from the graph database server system 120. The graph database server system 120 is configured to obtain the nodes and edges contained within the graph database(s), including data source devices 130, the graph database manipulation device 110, and any other source of graph data as appropriate to the requirements of specific applications in accordance with embodiments of the invention. The graph database contains a set of nodes and a set of edges representing the relationships between the nodes. In a variety of embodiments, nodes include node metadata indicating the content of the node and edges include edge weight metadata and edge display metadata. It should be noted that any of a variety of data could be incorporated into the nodes and/or edges as appropriate to the requirements of specific applications in accordance with embodiments of the invention. In a variety of embodiments, the graph database contains one or more references (such as a uniform resource locator) to nodes and/or edges that are stored in a distributed fashion across multiple systems, including third-party systems.
The graph database manipulation device 110 is further configured to generate a representation of the graph database from the perspective of a node within the graph database. The representation of the graph database includes a set of related nodes recursively identified based on the source node and the edges connecting the source node and the related nodes. In a variety of embodiments, the nodes and/or edges within the graph database have permission metadata determining which nodes and edges are available to be included within the perspective of the source node. The layout of the generated representation can be based on the edge weight metadata and the edge display metadata contained in the edges while the content of the generated representation can be based on node metadata contained in the nodes. In many embodiments, the graph database manipulation device 110 is configured to display a visualization of the generated representation using a display device. The visualization of the representation can be based on the source node, the edge weight metadata, and the edge display metadata in that a particular set of nodes can have a different context (and therefore a different visualized representation) based on the source node. In several embodiments, the generated representation approximates the nodes and/or edges contained in the graph database. Techniques for approximating a graph database from the perspective of a source node that can be utilized as appropriate to the requirements of specific applications in accordance with embodiments of the invention are described in more detail below.
A variety of manipulations can be performed on the generated representation using the graph database manipulation device 110 as appropriate to the requirements of specific applications in accordance with embodiments of the invention, including those described in more detail below. The graph database manipulation device 110 modifies the graph database based on the received manipulations, thereby recursively modifying the nodes and edges present in the graph database. In those embodiments where one or more nodes and/or edges are references to a data source device 130, the graph database manipulation device 110 is configured to transmit the modifications to the nodes and/or edges to the data source device 130 to be applied.
Graph database manipulation systems in accordance with embodiments of the invention are described above with respect to
In many cases, a representation of a graph database is generated and visualized to enable a user to explore the graph database. Graph database manipulation devices are configured to generate representations of a graph database from the perspective of one or more source nodes and manipulate the graph database based on modifications to the generated representations. A graph database manipulation device in accordance with an embodiment of the invention is conceptually illustrated in
The graph database manipulation application 232 configures processor 210 to perform a graph database manipulation processes based on the node data 234 and the edge data 236. In many embodiments, the node data 234 and the edge data 236 include the set of nodes and/or set of edges contained within a graph database (or a portion thereof). The graph database manipulation process includes determining one or more source nodes within the node data 234, recursively locating a set of related nodes in the node data 234 based on edges describing relationships between the nodes in the edge data 236, and generating a representation of the graph database from the perspective of the source node and including the related nodes. The layout of the nodes within the generated representation can be based on the edge weight metadata and the edge display metadata contained within the edges relating the nodes in the generated representation. Techniques for generating the representation of the nodes based on the edge weight metadata and the edge display metadata (including techniques for approximating the representation of the nodes and edges) that can be utilized in accordance with embodiments of the invention are described in more detail below. The generated representation of a node can include node metadata associated with the node describing the concept represented by the node within the graph database. In a variety of embodiments, the graph database manipulation device 200 includes a display device 240 connected to the processor 210 and configured to display a visualization of the generated representation. In many embodiments, the node data 234 and/or edge data 236 has permission metadata associated with the nodes and/or edges. The locations of related nodes and/or the generation of the representation of the located nodes are based on the permission metadata, where a node and/or edge cannot be included in the location of related nodes if the permission thresholds indicated in the permission metadata are not met.
The graph database manipulation process further includes receiving modifications to the generated representation. In several embodiments, the graph database manipulation device 200 includes an input device 242 configured to receive input indicative of modifications to the generated representation. The input device 242 is also configured to receive input indicative of a source node within a graph database. The input device 242 includes keyboards, mice, cameras, touch interfaces, and any other input device as appropriate to the requirements of specific applications in accordance with embodiments of the invention. Based on the modifications to the generated representation, the node data 234 and/or the edge data 236 are updated as appropriate to the received modifications. In embodiments where the node data 234 and/or edge data 236 are stored using a remote graph database server system, the graph database manipulation process configures the processor 210 to transmit the modifications using the network interface 220. When the nodes and/or edges include permission metadata, the manipulation of the node data 234 and/or edge data 236 (including the transmission of the manipulations) can be based on the permission metadata. For example, if a node has read-only permissions defined in the permission metadata, the graph database manipulation process could not allow modifications to the node data within the visualized representation and/or not update(or transmit) the node data 234 and/or the edge data 236 based on the received modifications.
Graph database manipulation devices in accordance with embodiments of the invention are described above with respect to
Graphs contain nodes and edges describing the relationships between the nodes. A graph-based representation of data can be explored by analyzing the nodes and the relationships between the nodes to gain insight into the underlying concepts and relationships. Graph databases in accordance with embodiments of the invention are configured to store conceptual data in nodes and the relationships between the nodes in the edges. Graph databases are utilized by graph database manipulation devices to visualize and manipulate the graph databases as appropriate to the requirements of specific applications in accordance with embodiments of the invention. A process for generating a graph database based on source data in accordance with an embodiment of the invention is conceptually illustrated in
The obtained (310) source data includes concepts and relationships between the concepts. Source data can be obtained (310) from any of a variety of data sources, such as data source devices, user input indicative of concepts and relationships, and any other data source as appropriate to the requirements of specific applications in accordance with embodiments of the invention. In many embodiments, the obtained (310) source data includes permission data describing the permissions associated with the concepts and/or relationships within the source data. The determined (312) associated data can include determining relationship weights and/or display data based on the obtained (310) source data. In several embodiments, the determined (312) associated data is included in the obtained (310) source data. The determined associated (312) data can also include determining relationships between concepts in the obtained (310) source data. In a number of embodiments, determining (312) associated data includes determining permission metadata based on the obtained (310) concepts, relationships, and/or the data source providing the source data. The permission metadata can include a security mechanism (e.g. a password, a security token, a handshake protocol, and/or a message) for accessing some or all of the data associated with a node, the neighbors associated with the node (e.g. nodes within a graph database having edges in common with the node), and/or payment thresholds as appropriate to the requirements of specific applications in accordance with embodiments of the invention. For example, a node can have permission metadata that provides a first set of node (meta)data for free, while if a payment is made the node will provide a second set of node (meta)data and/or edges indicating relationships to other nodes.
The determined (314) weights can be based on a variety of factors as appropriate to the requirements of specific applications in accordance with embodiments of the invention, such as the obtained concepts, the relationships between the concepts, and the determined (312) associated data. Encoding (316) source data includes encoding the relationship weights into edge weight metadata and/or encoding the display data into edge display metadata. In a number of embodiments, the relationships weights are encoded into edge weight metadata as a complex number (e.g. a number with a real component and an imaginary component) and the display data is encoded into edge display metadata as a binary string. It should be noted, however, that the encoding (316) of edge weight metadata and edge display metadata can be any format, including real numbers and character strings, as appropriate to the requirements of specific applications in accordance with embodiments of the invention. Furthermore, the edge display metadata and the edge weight metadata can be a single piece of metadata as appropriate to the requirements of specific applications of embodiments of the invention. The generated (318) node data includes concepts from the obtained (310) source data. In several embodiments, the generated (318) node data includes node metadata (including a string representing the concept represented by the node data) and references to one or more pieces of edge data. The character string can be taken from the obtained (310) source data and/or be an aggregation of related concepts within the source data. The generated (320) edge data includes the relationships contained within the obtained (310) source data and/or determined (312) associated data. A generated (320) piece of edge data includes the determined (314) edge weight metadata and the edge display metadata for the relationship used to generate the edge data. In a variety of embodiments, the generated (318) node data and/or the generated (320) edge data includes permission metadata. The generated (318) node data and/or the generated (320) edge data can also include latency metadata describing a time associated with retrieving data from the node and/or edge, such as a communication latency and/or a processing latency as appropriate to the requirements of specific applications in accordance with embodiments of the invention.
Although specific processes for the generation of graph databases are discussed above with respect to
In order to effectively explore a graph database, it is useful to generate a representation of the graph database from the perspective of a source node. This representation can then be visualized and explored to traverse the concepts and relationships contained within the graph database. Graph database manipulation devices in accordance with embodiments of the invention are configured to generate and visualize representations of graph databases. A process for visualizing a graph database in accordance with an embodiment of the invention is conceptually illustrated in
A source node can be obtained (410) from within the graph database. In many embodiments, the obtained (410) source node corresponds to a user of a graph database manipulation device, although any node within the graph database can be the obtained (410) source node as appropriate to the requirements of specific applications in accordance with embodiments of the invention. In several embodiments, obtaining (410) a source node includes obtaining a portion of the data (e.g. a summary and/or a preview) of the data associated with the source node within the graph database. In a variety of embodiments, the obtained (410) source node is a node not present within the graph database that is virtually linked to one or more nodes within the graph database. In several embodiments, the permissions for the source node are determined (412) based on permission metadata in the source node. In a number of embodiments, the permissions for the source node are determined (412) based on the graph database manipulation device visualizing the graph database; other permissions can be determined (412) as appropriate to the requirements of specific applications in accordance with embodiments of the invention.
The perspective of the source node is determined (414) by recursively traversing the graph database. A number of techniques can be utilized to traverse the graph in accordance with embodiments of the invention, including, but not limited to, depth-first search (DFS), breadth-first search (BFS), iterative deepening DFS, depth-limited DFS, lexicographic BFS, beam search, and best-first search. Other techniques, including path finding techniques, can be utilized as appropriate to the requirements of specific applications in accordance with embodiments of the invention. A variety of factors can be utilized to define (e.g. limit) the perspective of the source node, including a distance threshold for a related node to the source node (measured by the edges between the nodes in the graph database), a time threshold based on the time taken to traverse the graph database, and/or permission metadata associated with the nodes and/or edges within the graph database. From the perspective of a first node, nodes within one edge of the first node can be known as related nodes, while nodes in excess of one edge from the first node can be known as sub-related nodes. Node weights are determined (416) based on the edge weight metadata included in the edges connecting the related nodes to the source node. In a variety of embodiments, the node weight for a particular related node is determined (416) based on the node weight for its predecessor node (e.g. the node traversed prior to the particular related node during the determination (414) of the node perspective) and the edge weight metadata for the edge connecting the particular related node and its predecessor node. A node layout can be generated (418) based on the determined node weight and the edge display metadata for the edge connecting the particular related node and the predecessor node. In many embodiments, the node weight determines the size (e.g. area) of the node within the generated (418) layout and the edge display metadata determines the position of the nodes within the generated (418) layout. Visual characteristics for the layout (e.g. color, shading, shape) of the node within the generated (418) layout can be determined utilizing the edge weight metadata and/or the edge display metadata. In a number of embodiments, the visual characteristics are encoded in a binary string. In many embodiments, the visual characteristics are encoded as a complex number. The size, position, and/or visual characteristics of a node can be absolute (e.g. the same regardless of layout) or dependent on the source node and/or other related nodes within the generated representation. For example, the position of the generated (418) layout for a particular node can depend on the other nodes being (or previously) laid out. In many embodiments, the node layout depends on the node metadata for the node(s) being laid out. The above process repeats until no additional nodes exist (420) in the perspective of the obtained (410) source node.
Turning now to
Turning now to
Returning now to
In a variety of embodiments, one or more of the nodes in the generated (418) layout can include image data that can be displayed. Similarly, a subgraph of nodes can be interpreted as image data in the visual representation of a particular portion of the graph database. Turning now to
Specific examples of node layouts for a specific perspective from the perspective of a source node within a graph database is conceptually illustrated in
By exploring a representation of a graph database, a user can wish to add, modify, and/or remove data within the graph database and/or modify the layout of the generated representation. These modifications should be propagated to the graph database so that the new data is available to other users and/or in future explorations of the graph database. Graph database manipulation devices in accordance with embodiments of the invention are configured to modify graph databases based on manipulations of the representation of the graph database. A process for manipulating a graph database in accordance with an embodiment of the invention is conceptually illustrated in
A node update includes a target node and changed data metadata indicating the properties of the node and/or edges connected to the node to be updated. In several embodiments, a node update is obtained (510) from an input device included in a graph database manipulation device. The changed data metadata can include any update to the node metadata, edge weight metadata, and/or the edge display metadata as appropriate to the requirements of specific applications in accordance with embodiments of the invention. These updates include, but are not limited to, shifting the layout of nodes within a generated representation, transforming the size, rotation, and/or shape of nodes within a generated representation, changing the color of one or more nodes (e.g. data brushing), adding and/or removing edges between nodes within the graph database, adding and/or removing nodes within the graph database, modifying node metadata describing the concept associated with a node, and adding, modifying, or removing aggregations present in one or more nodes within the graph database. Other node updates, including combinations of the described node updates and other updated not specifically described, can be utilized as appropriate to the requirements of specific applications in accordance with embodiments of the invention. Node update data can also indicate to which node(s) the update should be applied; node updates can be applied to a single node and/or edge, the entire graph database, and/or one or more subgraphs within the graph database as appropriate to the requirements of specific applications in accordance with embodiments of the invention. In several embodiments, the node updates are only applied to a particular perspective of a node and/or the nodes available to a particular graph database manipulation device. Node updates can be customized and/or based on a template related to the particular properties of the nodes and/or edges being modified in the node update.
In a variety of embodiments, a node update is obtained (510) from one or more nodes within the graph database. Determining (512) linked nodes includes identifying the target node for the update and recursively identifying related nodes within the graph database utilizing techniques similar to those described above. In many embodiments, permission metadata is analyzed (514) to determine if a particular node and/or edge will perform and/or transmit the obtained (510) node update data. Updating (516) a node and/or edge includes adding, removing, and/or modifying metadata associated with the node and/or edge. In a number of embodiments, updating (516) a node includes modifying edge weight metadata for an edge connected to the node by computing a new complex number based on the previous edge weight metadata and the obtained (510) node update. In several embodiments, updating (516) a node includes modifying edge display metadata for an edge connected to the node by computing a new binary string based on the previous edge display metadata and the obtained (510) node update. The transmission (518) of node update notifications can be performed by passing messages between the nodes. In a variety of embodiments, node update notifications are transmitted (518) to nodes (or nodes having edges) to which the node update should be performed. The transmission (518) of node update notifications allows for the distributed (and parallel) processing of node updates across multiple nodes within the graph database, including those nodes that are stored in a distributed fashion. Transmitting (518) node update notifications can include a delay based on various locking techniques within distributed systems and/or latency associated with the transmission and/or application of the node updates. In many embodiments, the generated representation of the graph database is refreshed before, during, or after the transmission (518) of the node update notification(s).
Turning now to
Turning now to
Turning now to
Although specific examples of generated perspectives from the perspective of a source node within a graph database and updates to that perspective are conceptually illustrated in
Graph database with a high degree of dimensionality (e.g. a large number of nodes and edges) can provide a large amount of information that negatively influences the ability to explore the graph database and gain insights into the data. Similarly, graph databases with a high degree of dimensionality can exceed the computational power and/or storage of many computing devices. By generating an approximation to the graph database, the information stored in the graph database can be more easily processed, explored, and manipulated. Particular portions of the approximate database can be used to identify regions of interest within the full graph database for further exploration in more detail. Graph database manipulation devices in accordance with embodiments of the invention are configured to generate approximations of graph databases to facilitate the processing and exploration of the graph database. A process for approximating a graph database in accordance with an embodiment of the invention is conceptually illustrated in
In a variety of embodiments, a source node is obtained (610) utilizing processes similar to those described above. In several embodiments, determining (612) the perspective of the source node utilizes processes similar to those described above. Edges can be approximated (614) utilizing a variety of techniques as appropriate to the requirements of specific applications in accordance with embodiments of the invention. These techniques include, but are not limited to, random sampling of nodes within a particular distance from the obtained (610) source node, sampling nodes based on the number of edges connected to the nodes related to the source node, analyzing priority metadata and/or permission metadata associated with the nodes and/or edges, and aggregating nodes and/or edges to generate an approximate graph from the perspective of the obtained (610) source node. The aggregation of nodes and/or edges can also be performed based on the layout of the nodes within a generated representation of the source node and the related nodes within the graph database. For example, if the generated representation of a set of nodes would be too small for a user to effectively explore, the nodes can be aggregated so that useful information can still be analyzed by the user.
In many embodiments, edges are approximated (614) utilizing smart sampling techniques. Smart sampling techniques in accordance with embodiments of the invention include determining outlier nodes. Outlier nodes include nodes with few spatial neighbors, nodes with long edges, and/or nodes with high divergence. The length of an edge can be determined based on a variety of criteria, such as the distance between nodes within a visualized representation of the graph database, a latency associated with traversing the edge, edge weight metadata associated with the edge, and/or computed based on the edge weight metadata and/or the edge display metadata. In a variety of embodiments, the weight of an edge is determined by computing the squared complex weight based on the edge weight metadata. In several embodiments, edges can be incorporated into the graph database within a visualization using a variety of techniques. These techniques include, but are not limited to, preferentially sampling edges that will be drawn over empty space, preferentially sampling for edges between nodes that are separated by empty space, or any other techniques as appropriated to the requirements of specific applications of the invention. A heuristic algorithm can be utilized to determine the nodes to be sampled based on the properties of the node (e.g. the contents of the node and/or the time the node was last updated) and/or the edges connected to the node, such as the longest incoming or outgoing edge for a given node. In a number of embodiments, generating (616) the approximate view can be based on the approximated (614) edges and the obtained (610) source node utilizing processes similar to those described above. Other techniques for the smart sampling of nodes and edges to approximate (614) edges within a graph database from the perspective of a source node can be utilized as appropriate to the requirements of specific applications in accordance with embodiments of the invention.
In a variety of embodiments, a localized approximation (such as a Voronoi decomposition, or binary tree of spatial relationships between groups) can be utilized to approximate and/or cluster a subgraph in space, by partitioning the graph database into partial segments related within Cartesian space and/or the graph structure itself. Nodes and edges within a partition can be identified, selected, and manipulated by the user as a group. In one embodiment, a spatial cursor representing a user effector could quickly identify and select a spatial subsegment of a viewed subgraph, which could then be operated on as a group (moved in space or re-linked to another cluster), according to the wishes of the user. According to the wishes of the user, the selected spatial subsegment could also be refined by further manipulations, such as allowing the cursor to hover over the same spatial location.
In a variety of embodiments, a localized subset of the graph database can be employed as a localized approximation or projection of a graph database by partitioning the graph database into partial segments related within the graph database manipulation device. In a variety of embodiments, this type of approximation or clustering is done to maintaining the graph over disparate hardware. In a variety of embodiments, these segments can be identified via an automated mechanism for identifying minimal graph cuts for partitioning the graph, or simply via a threshold of locality around a source node. In a variety of embodiments, when a source node is selected within the graph database manipulation device, the nodes and edges within the partial segment of the graph containing the source node can be utilized by the graph database manipulation device as an approximation of the graph database for the purpose of display and further computation. As the partial segment is explored within the graph database manipulation device, the view and content of the partial segment can be changed to continue representing the graph locally.
Specific processes for the approximation of graph databases are discussed above with respect to
Nodes within a graph database can be configured to provide a variety of information based on requests for that information. This information can be taken from the node itself and/or from related nodes connected to the node. The information provided by a node can be influenced by the node requesting the information, providing a security mechanism for only providing data to those nodes with permissions to view the data. Graph database manipulation devices in accordance with embodiments of the invention are configured to transmit, process, and receive node messages. A process for processing node messages in accordance with an embodiment of the invention is conceptually illustrated in
A node message contains data request metadata indicating the desired data from the node receiving the node message. In a number of embodiments, the node message includes metadata identifying the node transmitting the node message and/or the intended node(s) to receive the message. In several embodiments, a node message is obtained (710) via a network connection and/or from an input device included in a graph database manipulation device. In a number of embodiments, a node message is obtained (710) from another node within the graph database. In many embodiments, node permissions are determined (712) based on permission metadata for the node obtaining (710) the node message. In a number of embodiments, the node permissions are determined (712) based on the graph database manipulation device visualizing the graph database; other node permissions can be determined (712) as appropriate to the requirements of specific applications in accordance with embodiments of the invention.
Processing (714) the node message includes determining the data requested within the node message based on the data request metadata. In a variety of embodiments, determining the requested data includes calculating additional data based on the data available to the node obtaining (710) the node message. The available data can include data present within the node and/or data present in other nodes connected to the node within the graph database. To retrieve data from other nodes within the graph database, additional node messages requesting the data from the other nodes can be created and transmitted by the node obtaining (710) the node message. In a number of embodiments, additional nodes are brought into the perspective of the node obtaining (710) the node message in order to request and receive the data needed to prepare the requested data. In many embodiments, a node message response including the requested data is created and transmitted (716). The node message response can also include metadata describing the sender and intended recipients of the node message response along with permission metadata describing access to the requested data. The transmission (716) of a node message response can be performed utilizing processes similar to those described above.
In a variety of embodiments, messages sent between nodes can be used as a method to augment a node's perspective. A source node within the graph, connected via edges to a field of additional successor nodes, requests from it successors data about their outgoing edges. If the criteria for permissions are met, according to the above descriptions for the interaction between messages and permissions metadata, successor nodes can transmit to the original node data about their own outgoing edges and successors, turning second-order successors into first-order successors. In this way, the perspective of a source node can grow, augmenting the domain over which it can have an effect on the larger graph.
Specific processes for processing node messages are discussed above with respect to
Graph database manipulation systems can be used to process graph databases into, and/or out of, alternate data structures. In some embodiments, tabular data structures can be extracted from graph databases. Extraction of tabular data structures can allow for exportation of graph databases, and/or application of various machine learning tools more appropriate to the particular data structure. For example, it may be beneficial to handle data in a spreadsheet, relational database, or any other tabular data storage method as appropriate to the requirements of a given application. Additionally, some machine learning algorithms are commonly used on matrix-like inputs, such as, but not limited to, classification, regression, and metric analysis.
Turning now to
Process 1400 further includes determining (1412) features. In numerous embodiments, features are the value corresponding to an observation-attribute pair. In other words, a feature can be any keyed numerical value associated with a particular observation. Features and/or observations can be represented as nodes within a graph database. Features can be encoded by edges between nodes, and/or lack thereof. For example, given a node n representing an observation, and a node m representing a feature, an edge between n and m can represent the encoded value for feature of m with respect to the observation of n. In some embodiments, the value of features can be encoded as numerical weight for an edge. However, the value of binary features can be represented by the presence/lack of an edge between nodes. In many embodiments, features can be used in a variety of ways, including, but not limited to specifying literal information about the observation, and/or as metadata about the observation.
Features can be imputed (1414) by graph database manipulation systems. If a value for a feature is unknown, the value of a feature can be imputed using machine learning, interpolation and/or extrapolation from a model, and/or any other heuristic as appropriate to the requirements of a given application. In a variety of embodiments, missing values within the graph database can be indicative of an imputable feature. In some embodiments, the lack of an edge between nodes is a basis for creating a new attribute. Lack of a value can indicate that the nominal value should be a null value. In a number of embodiments, missing values can be imputed using a model of an observation's attributes. For example, given two attributes x and y, such that there is a generative model p(x,y), and for observation i, xi is known but yi is missing, yi can be imputed as yi=argmaxy′p(xi,y′). However, any number of methods can be used to impute features as appropriate to the requirements of a given application. In many embodiments, a combination of different imputation methods are used. Further, imputation of features can occur at any point, and in some embodiments, features are imputed more than once.
A feature matrix can be generated (1416) from the graph database using a variety of methods. In many embodiments, a feature matrix is a tabular data structure. Tabular data structures are a form of data table where rows correspond to an observation, and columns correspond to features. However, any data structure can be used as appropriate to the requirements of a given application. In a variety of embodiments, tabular data structures can be generated by taking the graph where the nodes correspond each to either a row or column of the table, and using the adjacency matrix of the graph to populate the values in the data table. For example, a simple feature matrix can correspond to a bipartite subgraph and be represented as the adjacency matrix of the bipartite subgraph. The bipartite subgraph can be either undirected, or allow for edges only one direction. The nodes in one partite set can correspond to rows, and the nodes in the other partite set can correspond to columns. The adjacency matrix can be written in block-diagonal form with the feature matrix as one block and zeros elsewhere.
By way of further example, any arbitrary graph G can be represented as a feature matrix, with a |G|x|G| matrix, where each node in G has a corresponding row and a corresponding column. In this way, the matrix entries are the adjacency matrix of G. However, an arbitrary graph can be covered by a set of bipartite subgraphs, each of which corresponds to a data table. This can correspond to a relational set of tables, where the cross-references correspond to nodes present in at least one subgraph. Further, given that there is a natural homomorphism between a relational database and a graph database, any operation expressible in one language can translate to the other. For example, the generators of relational algebra correspond to operations on the graph. As such, graph operations can be done via manipulation graph database itself and/or via changes to graph topology.
In many embodiments, mappings (or relationships) between feature matrices and subgraphs allows for partite set(s) to be considered as a generalized observation. Applying calculations as though they are observations can allow for simply specified generic functions, which can reveal different types of information about the data. In a variety of embodiments, any two sets of nodes can be used to define a feature matrix by ignoring internal links within each set, treating each set as though they were bipartite. While numerous methods of generating feature matrices have been described above as examples, any number of methods for generating feature matrices can be used as appropriate to the requirements of a given application.
Machine learning can be performed (1418) on a feature matrix. Feature matrices can be inputs into machine learning algorithms, such as, but not limited to, linear/logistic regression, deep learning, dimensionality reduction, clustering, neural networks, support vector machines, and/or any other machine learning algorithm as appropriate to the requirements of a given application. Analytics can be performed on feature matrices, such as, but not limited to statistical summaries, frequency analysis, correlation analysis, predictive analytics, or any other analytical method as appropriate to the requirements of a given application. In some embodiments, feature matrices corresponding to a subset of the graph representing particular area of interest within the graph as a whole can be generated and/or provided. In some embodiments, extraction of features can be enhanced by implementing graphs or subgraphs as directed graphs. Imposing directionality can expedite the identification of maximally connected bipartite sets. In numerous embodiments, nodes corresponding to observations can be useful as training data for machine learning algorithms. Parent nodes of identified observations of interest can be used to construct partite sets, which in turn can be used to generate additional features and/or feature matrices. The feature matrix can be provided (1420) by the graph database manipulation system.
While numerous methods and systems for extracting tabular data structures have been described above with respect to
Just as tabular data structures can be extracted from graph databases, tabular data structures can be automatically imported into graph databases. In many embodiments, raw data formatted in tables are converted into low-schema graph structures by graph database manipulation systems. In a variety of embodiments, reverse mapping from low-schema graph to table relational database is performed.
Turning now to
Importation of tabular data structures can be extended to importation of relational databases. In many embodiments, importation of relational databases involves generating foreign keys that represent related ideas across tables. In numerous embodiments, foreign keys can be automatically fused (i.e. combined to form a new (foreign) key value). A single entry for a column type represented in multiple tables can correspond to a single node. Every unique node that corresponds to a row in each of the original tables can point to at least the single node. In a variety of embodiments, conflicts are resolved by inserting additional nodes, node properties, and/or edge properties.
In many embodiments, hierarchical data structures can be imported into graph databases using similar methods. Hierarchical structures can map natively onto trees, which is a type of graph. The tree can become a directed acyclic graph without the need for adding attributes to the graph. Attributes can be stored as single nodes, linked to, or linking to, nodes in which they appear in the original tree. Illustrations of importations and exports are illustrated in
While methods for importing tabular data structures are described above with respect to
Due to the sheer amount of data and relationships between the data, it can be difficult to visualize the data and/or particular relationships within the data as the amount of data present can be considered “noise” that obscures the “signal” representing the desired data. However, summary graphs can be generated in accordance with embodiments of the invention that can overcome this issue. Summary graphs generated by graph database manipulation systems can allow a user to easily understand a complex dataset. Methods of generating summary graphs can allow database manipulation systems to highlight important features and trends within data otherwise not easily identified. Given a data graph G, a summary subgraph G′ can be generated from G.
Turning now to
Further, a summary subgraph G′ can be generated (1616). In many embodiments, G′ is generated from the subset of nodes from G and R. In some embodiments, all edges in G′ are preserved. In a variety of embodiments, edges between at least one unequal pair (p,q) of nodes in G, with edge weight w, where w=f(p,q,R), where f is an arbitrary function. In some embodiments, f is a measure of pairwise relationship between p and q over R. However, f can be any number of functions, including, but not limited to, the correlation coefficient/mutual information of the attributes corresponding to p and q over the set R, a measure of Jaccard similarity of various relations in R of p and q, and/or any other arbitrary function as appropriate to the requirements of a given application.
In some embodiments, G′ is visualized (1618). In many embodiments, G′ is visualized using the methods for graph visualization described above. In visualizing G′, visual attributes of the marker for each node p (e.g. color, size, fill, etc.), can be used to indicate low-order statistics of the distribution over R of the attribute corresponding to p. In this way, underlying relationships between attributes can be visualized. In some embodiments, visualizations of G′ provide a complementary view of visualizations of G. Relationships between clusters of nodes in G can become visio-spatially obvious in light of G′. Relationships between clusters of nodes can be n-point correlations or any other relationship as appropriate to the requirements of a given application. Further, while moving up and down between levels of a graph, G′ can be used to preserve useful information from hidden levels.
Graph database manipulation systems can modify summary subgraphs in a similar manner as described above with respect to graph databases. In numerous embodiments, modifications to the visualization of G′, and/or modifications to G′ itself can be propagated to G. In this way, G′ can be used to guide manipulation of G. Summary subgraphs can be used as a substrate and seed for graphical analyses of graph databases. Graphical analysis can be applied to G′ to provide various representations of the graph. In many embodiments, representations of G′ do not require an information-destroying projection of the data onto a lower-dimensional substance, and/or the explicit calculation of an embedding. In some embodiments, force-directed graph viewing can be applied to G′. In this way, clusters of similar observations can be made visio-spatially apparent. This can facilitate cohort selection, as well as enhancing human understanding of the data.
Turning now to
While several processes and methods for generating summary subgraphs have been described above with respect to
The practices of feature construction and feature selection are useful in a variety of applications, including, but not limited to, machine learning, and data analysis. The complexity of purely data-driven approaches to feature construction and feature selection grow combinatorally in increasingly rich data, which can increase run-time, and/or the likelihood of the identification of spurious relationships. Feature construction and selection can be enhanced by the incorporation of human knowledge. The inclusion of exogenous prior knowledge can reduce the statistical penalty of a lower signal-to-noise ratio (SNR) for true relationships within the graph database.
Turning now to
Process 1800 further includes traversing (1814) the graph database to populate superfeatures. Because Fn can traverse the graph database, it can be arbitrarily complex. In some embodiments, Fn is a wi-weighted average of a standardized X(e,mi), logical AND or logical OR of the X(e,mi), or any other arbitrarily complex function as appropriate to the requirements of a given application. In this way, features can be constructed by dynamically pulling information through the node. In a variety of embodiments, the construction of features is analogous to a data flow diagram. The computation of X(e,n) can be performed recursively. In some embodiments, limitations are placed on the recursion depth. The value of X(e,n) can be calculated on request and cached, calculated in the background, and/or dynamically when constructing superfeatures. Utilization of request and cache and background calculations can improve runtime performance in many embodiments of the invention. To avoid stale cached values, dirty bits can be used to denote portions of the underlying graph that have been manipulated. However, any synchronization method, including, but not limited to, static and/or dynamic incremental computing procedures, can be used as appropriate to the requirements of a given application.
Arbitrary, compound features can allow for structures to be imposed on graphs that would be impossible (or statistically difficult) to infer purely from data. For example, the distinction between certain categories may not be meaningful for a particular analysis, particularly in multifaceted datasets. Aggregating categories using superfeatures can increase SNR for relationships of interest. By way of example, if specific numeral values of a feature are less meaningful than whether the value hits a threshold or falls within a range, then a superfeature can be used to abstract the relationship. This method can be useful in many fields, including, but not limited to, the healthcare field where diagnostics with critical values can be an important indicator to monitor.
Superfeatures can also be used to enhance machine learning outcomes. For example, logical exclusive-or (XOR) can be difficult for machine learning algorithms to detect. In an unconditioned dataset, there may be separate categories that mean the same thing, e.g. “bachelor” and “unmarried.” A superfeature can be used to aggregate the categories. This can useful in a variety of contexts, such as enhancing understanding of a dataset. In a variety of embodiments, graph database manipulation systems can identify and/or flag potentially synonymous category labels. In some embodiments, synonymous category labels are detected using machine learning algorithms. In a variety of embodiments, dictionary and thesaurus datasets are used to flag category labels. Lexicographical analysis over the graph database, and/or other multiple datasets can be used to generate labels. However, any number of category detection methods can be used as appropriate to the requirements of a given application. Any graph that uses edges to quantify associations and/or relationships between concepts can be used as a basis to build superfeatures. Categories and/or relationships can be visualized as a data flow/knowledge graph and used to identify and link already-known relationships during data exploration.
In many embodiments, graph database manipulation systems can automatically generate and/or recommend superfeatures. In some embodiments, clusters of nodes can be collected together as superfeatures. In hierarchical structures (e.g. polytrees) that capture iterative subdivisions of a set, building superfeatures that compute the logical OR of an inner node's children nodes allows the choice of a particular partition of the set. Further, some superfeatures can be made “fuzzy,” by considering a weighted combination of features. In many embodiments, a fuzzy superfeature node nf has an assigned edge weight X(nf,n)=1 (where the notation e refers to the weight of the edge from the node in the first argument to the second) and X(nf,m)=g(X(n,m1), . . . X(n,m|G′|)) for all mi in G′, and some function g. By changing the choice of g, moments, kernels, and other functions can be input, allowing for the identification of specific patterns of interest. This can further allow for the summarization of heterogeneous data in tabular data structures. However, superfeatures are not the only type of supernode that graph database manipulation systems can generate.
Superobservations can be generated in a similar way to superfeatures. Superobservation nodes represent an aggregate over a group of observations. For example, a supernode s that is associated with observation nodes m1 . . . mk with corresponding weights w1 . . . wk, is calculated as: X(n,s)=Fs((w1,X(n,m1)), . . . (wk,X(mk,wk))). In this way, nodes m1 . . . mk are grouped together as a category. F can be any function, for example, a wi-weighted mean of X(n,mi). Superobservations can be visualized on a graph as either a node s with wi-weighted edges outgoing to mk, and/or with wi-weighted edges coming in to s. However, in many embodiments, there is no edge directionality. A similar effect to a superobservation can be achieved using a summary subgraph over {n,s}, and/or via a subgraph over {m1, m2, . . . , s1, s2, . . . }.
Further, superobservations can be computed recursively. A supernode can aggregate information from their children, parents, and/or neighbor nodes in an arbitrary way and similarly pass information through. Recursive computation of supernodes allows them to be more transparent and intuitive, as data can be visually traced along paths on the graph. Graph database manipulation systems can then, given a generalized feature node n, and a generalized observation node m on the graph G, associate the value X(n,m)=f(p1, . . . , pk), where pi are graph paths from node n to node m, for some arbitrary function f. Following, the paths are those that contain supernodes, and f is composed of the local functions corresponding to supernodes along the path. For example, X(n,m) can be the sum or maximum value of all paths between n and m, where the path is the product of its edge weights. Strength of edge weights can encode the strength of similarity (ontological) and/or the degree of certainty about similarity (epistemic). However, X(n,m) can be any value of interest as appropriate to the requirements of a given application.
While numerous methods for generating supernodes have been described above with respect to
Simple observation of a graph does not always lead to easy comprehension by a human observer. In many embodiments, graphs may be highly dimensional, noisy, or otherwise difficult for a human to interpret. Graph database manipulation systems can be used to automatically provide an explanation for a selected subset of observations. Explanatory variables refer to the predictive power of a given feature to a reference feature.
Turning now to
The set of explanatory variables can be expanded to include further unknown explanatory variables by application of machine learning and/or inference methods on the residual subgraph and/or feature space. The predictive power of an explanatory variable can be determined using a binomial classifier algorithm. However, this can be generalized to a higher-dimensional classifier algorithm, and/or an unsupervised learning algorithm. By way of example, let C={c1,c2, . . . , cm} denote the relevant feature nodes, and let C′ be the subset of C representing the features whose explanatory power is known. Denote Mx(A) as the feature matrix for the features in the set A over the relevant set of observations X. The vector of true observation labels is written as Y=Mx({y}). C′ can be partialized out of Y. In some embodiments, C′ is partialized by defining partialized by defining Y′ as a vector of probabilities of each observation being in class 1 rather than class 0, as per a trained classifier, and the residual Y-Y′ can be used as the dependent variable to train a regression or other algorithm on the feature matrix of the remaining variables Mx(C\C′). However, any number of partialization methods can be used as appropriate to the requirements of a given application. In this method, the strengths of the unknown relationship between the variables in C\C′ are given by the coefficients of this regression against the C′-partialized target variable.
While a specific method is described above as an example of identifying unknown explanatory variables, any number of various algorithms can be applied to achieve similar results in accordance with various embodiments of the invention. For example, a multivariate predictive model can provide a prediction of residual Y-Y′. Once unknown explanatory variables are identified, they can be encoded directly onto the graph, and/or provided as a separate structure. Identified explanatory variables can be useful in understanding large datasets. Manipulation of datasets can be used to further explore and elucidate patters and relationships within data. Graph database manipulation systems can perform a number of relational algebra operations in order to query and/or restructure graph databases. Methods for performing relational algebra operations are described below.
Relational algebra operations can be used to create relational database query languages. In many embodiments, queries are performed of the graph data using the graph data manipulation device in combination with the properties of the graph itself. In many embodiments, graph database manipulation systems allow for automatic application of operations to the graph database. Graph queries executed using the graph manipulation device can be equivalent to, and/or be more general than, relational database queries. In some embodiments of the graph manipulation device operating on directed weighted graphs. Actions can be equivalent to particular relational database query types defined by the relational algebra. In numerous embodiments, graph databases are visualized in windows. Operations can be applied to graph databases automatically via graphical manipulation of the graph within, and/or between windows. While operations are discussed below in the context of windows, windows are not necessary for the performance of operations using a graph database manipulation system.
Turning now to
Using the above described 7 operations fundamental to a relational algebra, a full relational algebra is accomplished utilizing 5 operations on a directed weighted graph. In many embodiments, a graph manipulation device utilizes these operations to perform equivalent to, and/or more general than, a full database query language. In a variety of embodiments, the graph database manipulation system uses spatial and/or numerical operations rather than sequential commands entered at a command line. The graph manipulation device can be configured to execute these operations automatically, allowing the equivalent of sequences of commands to be executed, albeit created without requiring the programming of textual computer code.
Although specific examples of directed weighted graph manipulation operations which are equivalent to a full relational algebra are illustrated in
Although the present invention has been described in certain specific aspects, many additional modifications and variations would be apparent to those skilled in the art. In particular, any of the various processes described above can be performed in alternative sequences and/or in parallel (on the same or on different computing devices) in order to achieve similar results in a manner that is more appropriate to the requirements of a specific application. It is therefore to be understood that the present invention can be practiced otherwise than specifically described without departing from the scope and spirit of the present invention. Thus, embodiments of the present invention should be considered in all respects as illustrative and not restrictive. It will be evident to the person skilled in the art to freely combine several or all of the embodiments discussed here as deemed suitable for a specific application of the invention. Throughout this disclosure, terms like “advantageous”, “exemplary” or “preferred” indicate elements or dimensions which are particularly suitable (but not essential) to the invention or an embodiment thereof, and may be modified wherever deemed suitable by the skilled person, except where expressly required. Accordingly, the scope of the invention should be determined not by the embodiments illustrated, but by the appended claims and their equivalents.
The current application is a continuation of U.S. patent application Ser. No. 15/493,499, filed Apr. 21, 2017 and issued as U.S. Pat. No. 10,776,965 on Sep. 15, 2020, which is a continuation-in-part of U.S. patent application Ser. No. 15/136,426, filed Apr. 22, 2016 and issued as U.S. Pat. No. 9,740,744 on Aug. 22, 2017, which is a continuation of U.S. patent application Ser. No. 14/318,432, filed Jun. 27, 2014 and issued as U.S. Pat. No. 9,348,947 on May 24, 2016, which claims priority to U.S. Provisional Patent Application No. 61/858,782, filed Jul. 26, 2013. The current application also claims priority to U.S. Provisional Patent Application No. 62/325,879, filed Apr. 21, 2016. The disclosures of U.S. patent application Ser. Nos. 15/493,499, 15/136,426, 14/318,432 and U.S. Provisional Patent Application Nos. 61/858,782 and 62/325,879 are hereby incorporated by reference in their entireties.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5844572 | Schott | Dec 1998 | A |
| 7539697 | Akella et al. | May 2009 | B1 |
| 7644052 | Chang et al. | Jan 2010 | B1 |
| 7865534 | Chandra et al. | Jan 2011 | B2 |
| 8185558 | Narayanan et al. | May 2012 | B1 |
| 8352465 | Jing et al. | Jan 2013 | B1 |
| 8572129 | Lee | Oct 2013 | B1 |
| 8736612 | Goldman | May 2014 | B1 |
| 8739016 | Goldman | May 2014 | B1 |
| 9129158 | Medasani et al. | Sep 2015 | B1 |
| 9187088 | Ferguson et al. | Nov 2015 | B1 |
| 9348947 | Stetson et al. | May 2016 | B2 |
| 9411890 | Komarov | Aug 2016 | B2 |
| 9684678 | Hatami-hanza | Jun 2017 | B2 |
| 9740744 | Stetson et al. | Aug 2017 | B2 |
| 9792530 | Wu et al. | Oct 2017 | B1 |
| 10019011 | Green et al. | Jul 2018 | B1 |
| 10713258 | Stetson et al. | Jul 2020 | B2 |
| 10776965 | Stetson et al. | Sep 2020 | B2 |
| 11507099 | Stetson et al. | Nov 2022 | B2 |
| 20020087275 | Kim et al. | Jul 2002 | A1 |
| 20040034848 | Moore | Feb 2004 | A1 |
| 20070124291 | Hassan et al. | May 2007 | A1 |
| 20070208693 | Chang et al. | Sep 2007 | A1 |
| 20070239694 | Singh | Oct 2007 | A1 |
| 20070294221 | Chen | Dec 2007 | A1 |
| 20080059512 | Roitblat et al. | Mar 2008 | A1 |
| 20080097941 | Agarwal | Apr 2008 | A1 |
| 20090013314 | Opaterny | Jan 2009 | A1 |
| 20090303239 | Ang | Dec 2009 | A1 |
| 20100094864 | Foody | Apr 2010 | A1 |
| 20100274815 | Vanasco | Oct 2010 | A1 |
| 20110173189 | Singh et al. | Jul 2011 | A1 |
| 20110249002 | Duplessis | Oct 2011 | A1 |
| 20110295841 | Sityon | Dec 2011 | A1 |
| 20120221555 | Byrne et al. | Aug 2012 | A1 |
| 20120293542 | Iwama | Nov 2012 | A1 |
| 20120317149 | Jagota | Dec 2012 | A1 |
| 20130117261 | Sambrani | May 2013 | A1 |
| 20130159901 | Wolge | Jun 2013 | A1 |
| 20130191416 | Lee | Jul 2013 | A1 |
| 20130246342 | Faith | Sep 2013 | A1 |
| 20130290234 | Harris | Oct 2013 | A1 |
| 20140149376 | Kutaragi et al. | May 2014 | A1 |
| 20140172914 | Elnikety | Jun 2014 | A1 |
| 20150033106 | Stetson et al. | Jan 2015 | A1 |
| 20150254331 | Long et al. | Sep 2015 | A1 |
| 20160239545 | Stetson et al. | Aug 2016 | A1 |
| 20170131719 | Micks et al. | May 2017 | A1 |
| 20170221240 | Stetson et al. | Aug 2017 | A1 |
| 20170316059 | Stetson et al. | Nov 2017 | A1 |
| 20180014217 | Kleinbeck et al. | Jan 2018 | A1 |
| 20180189634 | Abdelaziz et al. | Jul 2018 | A1 |
| 20190025841 | Haynes et al. | Jan 2019 | A1 |
| 20190310650 | Halder | Oct 2019 | A1 |
| 20190384863 | Sirin et al. | Dec 2019 | A1 |
| 20200081445 | Stetson et al. | Mar 2020 | A1 |
| 20230202513 | Stetson et al. | Jun 2023 | A1 |
| Number | Date | Country |
|---|---|---|
| 3850546 | Jul 2021 | EP |
| 2018010434 | Jan 2018 | WO |
| 2020055910 | Mar 2020 | WO |
| Entry |
|---|
| International Preliminary Report on Patentability for International Application No. PCT/US2019/050469, Issued Mar. 9, 2021, Mailed Mar. 25, 2021, 7 Pgs. |
| International Search Report and Written Opinion for International Application No. PCT/US2019/050469, Search completed Nov. 20, 2019, Mailed Dec. 11, 2019, 9 Pgs. |
| Madrigal, “Inside Waymo's Secret World of Training Self-Driving Cars”, The Atlantic, Aug. 23, 2017, Retrieved from: https://www.theatlantic.com/technology/archive/2017/08/inside-waymos-secret-testing-and-simulation-facilities/537648/, 26 pgs. |
| Meyerovich et al., “Superconductor: A Language for Big Data Visualization”, ACM, LASH-C, Shenzen, China, Feb. 24, 2013. |
| Monti et al., “Geometric Deep Learning on Graphs and Manifolds Using Mixture Model CNNs”, In: Cornell University Library/ Computer Science/Computer Vision and Pattern Recognition, Nov. 25, 2016, (online) (retrieved on Nov. 19, 2019 (Nov. 19, 2019). |
| Torok, Matthew E., “Superconductor 1 Million Nodes at 30 FPS”, Retrieved from: http://parlab.eecs.berkeley.edu/sites/all/parlab/files/superconductor%20overview%20external.pdf, Aug. 2012, 29 pgs. |
| Extended European Search Report for European Application No. 19860523.0, Search completed May 5, 2022, Mailed May 13, 2022, 16 Pgs. |
| International Preliminary Report on Patentability for International Application PCT/US2019/050469, Report issued Mar. 9, 2021, Mailed Mar. 25, 2021, 7 Pgs. |
| “Relational algebra”, Wikipedia, Retrieved from: https://en.wikipedia.org/wiki/Relational_algebra, Last modified Jul. 28, 2022, Printed Sep. 19, 2022, 11 pgs. |
| Bagschik et al., “Ontology Based Scene Creation for the Development of Automated Vehicles”, IEEE Intelligent Vehicles Symposium (IV), IEEE, Jun. 26, 2018, pp. 1813-1820, XP033423512. |
| Chauvin, “Hierarchical Decision-Making for Autonomous Driving”, Aug. 2018, pp. 1-8, XP055916113, DOI: 10.13140/rg.2.2.24352.43526. |
| Chen et al., “Manifold Learning-Based Data Sampling for Model Training”, Informatik Aktuell, Feb. 21, 2018, Springer Verlag, London., GB, pp. 269-274, XP055916157. |
| Codd, E.F., “A Relational Model of Data for Large Shared Data Banks”, Communications of the ACM, vol. 13, No. 6, Jun. 1970, pp. 377-387. |
| Dopfer et al., “What can we learn from accident videos?”, 2013 CACS International Automatic Control Conference (CACS), IEEE, Dec. 2, 2013, pp. 68-73, XP032563875, DOI: 10.1109/CACS.2013.6734109. |
| Guo et al., “Semantically Smooth Knowledge Graph Embedding”, Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing Jul. 26, 2015, vol. 1, pp. 84-94, XP055757075. |
| Olsson, “Behavior Trees for decision-making in Autonomous Driving”, 2016, retrieved from http://www.diva-portal.org/smash/get/diva2:907048/FULLTEXT01.pdf (Year: 2016). |
| Wilcke et al., “The knowledge graph as the default data model for learning on heterogeneous knowledge”, Data Science, Dec. 8, 2017, vol. 1, No. 1-2, pp. 39-57, XP055915793. |
| Number | Date | Country | |
|---|---|---|---|
| 20210065415 A1 | Mar 2021 | US |
| Number | Date | Country | |
|---|---|---|---|
| 62325879 | Apr 2016 | US | |
| 61858782 | Jul 2013 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 15493499 | Apr 2017 | US |
| Child | 17017541 | US | |
| Parent | 14318432 | Jun 2014 | US |
| Child | 15136426 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 15136426 | Apr 2016 | US |
| Child | 15493499 | US |
