Patent Application
20120265784
Typically, search engines provide the user with an empty search box into which a user may input a search query and interact with the search system. For various reasons, search queries provided by users do not always accurately reflect the user's actual intent. While some search engines are capable of assisting users in completing search queries, e.g., by providing query completion suggestions, such query completion suggestions are typically generated based solely upon search queries that have been submitted in the past, such as the most popular search queries that begin with the alphanumeric characters with which the user begins their query. As such, if a user begins entering the search query “wate,” query completion suggestions may include “water for elephants,” “water fountain,” “water cycle,” “watergate,” etc. However, in many cases, the suggestions provided to the user may not reflect what the user is actually intending to search for as they are based solely on prior queries input by a collective group of users, and do not include unseen queries or queries that are seldom entered, but that may be what the user is intending to search for. Likewise, users have little, if any, knowledge about what information the search system may have available with regard to answering the task they have in mind. As such, the search system may not return satisfactory results to the user due, at least in part, to the system's failure to understand or disambiguate the user's actual intent, and to match the user's intent with useful information.
This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.
Embodiments of the present invention relate to ordering or ranking semantically-identified query formulation suggestions. Semantic query patterns are identified for a plurality of search queries and a weight is identified for each. Also identified are a plurality of semantic categories, and each also has a weight identified therefore. Additionally, terms or phrases commonly associated with the semantic categories are identified, as are semantic attributes as they pertain to the semantic categories. Semantic attribute patterns and respective weights therefore are then identified. A text-parser is then generated or customized from the semantic query patterns and respective weights, the semantic category terms, and the semantic attribute patterns and respective weights, the text-parser for use in parsing input user queries or portions thereof. When a user search query is received, the text-parser is applied to determine at least one likely attribute, attribute value, term commonly associated with a semantic category, or a combination thereof and the determined attribute/attribute value/term is transmitted for presentation with an order or ranking representative of the respective calculated weights.
The present invention is described in detail below with reference to the attached drawing figures, wherein:
The subject matter of the present invention is described with specificity herein to meet statutory requirements. However, the description itself is not intended to limit the scope of this patent. Rather, the inventors have contemplated that the claimed subject matter might also be embodied in other ways, to include different steps or combinations of steps similar to the ones described in this document, in conjunction with other present or future technologies. Moreover, although the terms “step” and/or “block” may be used herein to connote different elements of methods employed, the terms should not be interpreted as implying any particular order among or between various steps herein disclosed unless and except when the order of individual steps is explicitly described.
Embodiments of the present invention provide for systems, methods and computer-storage media having computer-usable instructions embodied thereon, for building, linking, and exposing semantic knowledge of a search system or systems (e.g., query logs, facets, relationships of entities from structured and unstructured data, contextual signals, and the like) to assist users in query formulation and intent disambiguation. Embodiments of the invention may expose never-seen queries and provide better user-intent signals to the system. Such signals lead to selection and presentation of better and more relevant search results.
Accordingly, in one embodiment, the present invention is directed to a method performed by a computing device having a processor and a memory for identifying query formulation suggestions in response to receiving a search query. The method includes receiving at least a portion of the search query and identifying one or more query formulation suggestions by semantically analyzing the at least the portion of the search query. The one or more query formulation suggestions comprise semantic-pattern-based query suggestions that are derived from semantic query patterns and at least one of one or more entities or information associated with the one or more entities. The method further includes transmitting the one or more query formulation suggestions for presentation.
In another embodiment, the invention is directed to a system comprising a processor and a memory for identifying query formulation suggestions in response to receiving a search query. The system includes a receiving component, a semantic analyzing component and a transmitting component. The receiving component receives at least a portion of a search query. The semantic analyzing component semantically analyzes the at least the portion of the search query to identify one or more query formulation suggestions that comprise semantic-pattern-based query suggestions that are derived from semantic query patterns and at least one of one or more entities or information associated with the one or more entities. The transmitting component transmits the one or more query formulation suggestions for presentation.
A further embodiment of the invention is directed to one or more computer-storage media storing computer-useable instructions that, when used by a computing device, cause the computing device to perform a method for identifying query formulation suggestions in response to receiving a search query. The method includes receiving at least a portion of the search query and, contemporaneous to a formulation of the at least the portion of the search query, identifying one or more query formulation suggestions by semantically analyzing the at least the portion of the search query. The one or more query formulation suggestions comprise query-log-based query suggestions and semantic-pattern-based query suggestions that are derived from semantic query patterns and at least one of one or more entities or information associated with the one or more entities. The information comprises one or more attributes and one or more attribute values. The method further includes transmitting the query formulation suggestions for presentation, wherein the query formulation suggestions are configured to be selectable by a user to formulate the search query. Still further, the method includes receiving a user selection of one of the query formulation suggestions, and modifying the at least the portion of the search query according to the selected query formulation suggestion.
Yet another embodiment of the invention is directed to a method performed by a computing device having a processor and memory for identifying query formulation suggestions. The method includes receiving at least a portion of a search query and identifying one or more semantic-pattern-based query suggestions by semantically analyzing the at least the portion of the search query. The one or more semantic-pattern-based query suggestions are derived from semantic query patterns, one or more entities, and information associated with the one or more entities, and the information includes attributes and attribute values such that the one or more entities are used to identify the one or more semantic-pattern-based query suggestions. The method further includes identifying a set of the query formulation suggestions to transmit for presentation, and transmitting the set of query formulation suggestions for presentation. The one or more query formulation suggestions are configured to be selectable by a user to formulate the search query.
In another embodiment, the invention is directed to one or more computer-storage media storing computer-useable instructions that, when used by a computing device, cause the computing device to perform a method for identifying query formulation suggestions. The method includes receiving at least a portion of a search query and semantically analyzing the at least the portion of the search query to identify one or more semantic-pattern-based query suggestions that are generated by at least one of semantic query patterns, one or more entities, and information associated with the one or more entities. The method further includes transmitting the one or more semantic-pattern-based query suggestions for presentation. The one or more semantic-pattern-based query suggestions being configured to be selectable by a user to formulate the search query. Still further, the method includes receiving a user selection of one of the one or more semantic-pattern-based query suggestions. In response to the user selection, the method additionally includes modifying the at least the portion of the search query based on the selected semantic-pattern-based query suggestion. Further, the method includes updating a set of search results based on the modified search query.
In another embodiment, the invention is directed to a user interface embodied on one or more computer-storage media, the user interface for presenting query formulation suggestions. The user interface includes a query display area, an attribute suggestion presenting area and an attribute value suggestion presenting area. The query display area is configured for displaying at least a portion of a search query received from a user, wherein the at least the portion of the search query is semantically analyzed to identify one or more semantic-pattern-based query suggestions by semantically analyzing the at least the portion of the search query, and wherein the one or more semantic-pattern-based query suggestions includes at least one of one or more entities, one or more attributes, or one or more attribute values. The attribute suggestion presenting area is configured for presenting a list of attributes as query formulation suggestions, each attribute in the list of attributes being associated with at least one of the one or more entities. The attribute value suggestion presenting area is configured for presenting a list of attribute values as the query formulation suggestions upon receiving a user selection of one of the attributes, each attribute value in the list of attribute values being associated with the selected attribute. The attribute value suggestion presenting area is arranged relative to the attribute suggestion presenting area such that the attribute values associated with a particular one of the attributes are presented in proximal association with their respective corresponding attributes.
In another embodiment, the invention is directed to one or more computer-storage media storing computer-useable instructions that, when used by a computing device, cause the computing device to perform a method for generating query formulation suggestions. The method includes receiving a plurality of search queries from a query log, each of the plurality of search queries being associated with a respective set of matching entities, and identifying at least one semantic query pattern from the plurality of search queries. The method further includes identifying a weight for each identified semantic query pattern; identifying a plurality of semantic categories from an entity database, each of the plurality of semantic categories being associated with a respective set of entities; identifying at least one term or phrase (e.g, an n-gram) commonly associated with at least one of the semantic categories; identifying a plurality of semantic attributes as they pertain to the plurality of semantic categories; identifying at least one semantic attribute pattern from the plurality of semantic attributes; identifying a weight for each identified semantic attribute pattern; and generating a text-parser from the at least one semantic query pattern and respective weights, the at least one semantic category term or phrase and respective weights, and the at least one semantic attribute patterns and respective weights. The text parser is configured for use in parsing input user queries or portions thereof.
Having briefly described an overview of embodiments of the present invention, an exemplary operating environment in which embodiments of the present invention may be implemented is described below in order to provide a general context for various aspects of the present invention. Referring initially to
Embodiments of the invention may be described in the general context of computer code or machine-useable instructions, including computer-executable instructions such as program modules, being executed by a computer or other machine, such as a personal data assistant or other handheld device. Generally, program modules including routines, programs, objects, components, data structures, etc., refer to code that performs particular tasks or implement particular abstract data types. Embodiments of the invention may be practiced in a variety of system configurations, including handheld devices, consumer electronics, general-purpose computers, more specialty computing devices, and the like. Embodiments of the invention may also be practiced in distributed computing environments where tasks are performed by remote-processing devices that are linked through a communications network.
With reference to
The computing device 100 typically includes a variety of computer-readable media. Computer-readable media can be any available media that can be accessed by the computing device 100 and includes both volatile and nonvolatile media, removable and non-removable media. By way of example, and not limitation, computer-readable media may comprise computer-storage media and communication media. Computer-storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data. Computer-storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by the computing device 100. Communication media typically embodies computer-readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media. The term “modulated data signal” means a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared and other wireless media. Combinations of any of the above should also be included within the scope of computer-readable media.
The memory 112 includes computer-storage media in the form of volatile and/or nonvolatile memory. The memory may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical-disc drives, and the like. The computing device 100 includes one or more processors that read data from various entities such as the memory 112 or I/O components 120. Presentation component(s) 116 present data indications to a user or other device. Exemplary presentation components include a display device, speaker, printing component, vibrating component, etc.
The I/O ports 118 allow the computing device 100 to be logically coupled to other devices including I/O components 120, some of which may be built in. Illustrative components include a microphone, joystick, game pad, satellite dish, scanner, printer, wireless device, and the like.
Referring now to
In some embodiments, one or more of the illustrated components/modules may be implemented as stand-alone applications. In other embodiments, one or more of the illustrated components/modules may be integrated directly into other components. It will be understood by those of ordinary skill in the art that the components/modules illustrated in
It should be understood that this and other arrangements described herein are set forth only as examples. Other arrangements and elements (e.g., machines, interfaces, functions, orders, and groupings of functions, etc.) can be used in addition to or instead of those shown, and some elements may be omitted altogether. Further, many of the elements described herein are functional entities that may be implemented as discrete or distributed components or in conjunction with other components/modules, and in any suitable combination and location. Various functions described herein as being performed by one or more entities may be carried out by hardware, firmware, and/or software. For instance, various functions may be carried out by a processor executing instructions stored in memory.
The entity store 206 is configured to store various types of information used by the semantic suggestion generator 208 to identify query formulation suggestions based on a semantic analysis of at least the portion of the search query entered by the user. The type of information stored in the entity store 206 may include, for example, entities, categories, attributes, and attribute values, as more fully described below. More specifically, in various embodiments, the information stored in the entity store 206 may include, without limitation, entities or items (e.g., products available for commerce, locations, people, companies, organizations, and the like), categories to which such entities may correspond, attributes associated with the entities, and attribute values associated with the entities as they pertain to particular attributes thereof. The information in the entity store 206 may also include common queries that previously have been issued or submitted to the search engine 204, in addition to popular or frequently selected interpretations. In an embodiment, at least a portion of these common queries may be stored in a query log such that the query log may be accessed to identify common queries.
The information stored in the entity store 206 may be used as query formulation suggestions. Query formulation suggestions, as used herein, refer to information that can be used to discover a user's true intent by assisting the user to formulate a query, for instance, by expanding upon or clarifying a partial query that has been entered by the user. As such, in one embodiment, query formulation suggestions are first presented prior to the determination and/or presentation of any search results determined to satisfy the query (or partial query). As will be discussed in further detail, in embodiments hereof, search results may be determined and presented when an input search query, or portion thereof, reaches a semantic boundary. As such, if a partial search query is entered, query formulation suggestions may be presented before search results are even identified. However, once the input query is modified to the point that a semantic boundary is reached, search results may be determined and presented. One or more different or updated query formulation suggestions may then be presented based on the modified search query. Thus, those of skill in the art will understand that query formulation suggestions differ from search results not only in content (although in some instances the content may be substantially similar), but also in their use upon user selection thereof. Selection and use of query formulation suggestions is more fully described below.
Query formulation suggestions may include various types of suggestions for formulating queries including, by way of example and not limitation, query-log-based query suggestions, semantic-pattern-based query suggestions (e.g., template-based query suggestions), categories associated with a search query, attributes associated with one or more identified entities, and attribute values associated with one or more identified entities. It should be noted that query formulation suggestions are generally identified, presented, and selected contemporaneously with query formulation, not post-query submission.
“Query-log-based query suggestions” are suggestions that attempt to aid the user in formulation of a search query by providing the user with the most popular previously-submitted search queries identified from a query log that correspond to the character sequence entered into the search box at a particular instance. “Semantic-pattern-based query suggestions,” on the other hand, refer to query formulation suggestions that are generated from semantic query patterns (e.g., templates). When a search query, or portion thereof, is submitted to the search system, a database of semantic query patterns (such as “<brand>camera,” as more fully described below) is accessed and an attempt is made to identify one or more semantic query patterns that correspond in some way to the input search query. In generating semantic-pattern-based query suggestions, a database or entity store 206 may be used. By accessing entities and their respective associated information from the entity store 206, it can be ensured that only those semantic-pattern-based query suggestions that match known entities are returned. As such, before being presented to a user, semantic-pattern-based query suggestions are compared to entities stored in the entity store 206. In one embodiment, if a particular semantic-pattern-based query suggestion is not found in the entity store 206, it is not returned to the user as a query suggestion, as it may not exist (e.g., such as a particular product that does not exist), or at least it does not match with the known information in the entity store 206. But if it is found, it may be returned to the user, as further discussed herein. Semantic-pattern-based query suggestions may be presented to the user in much the same way as query-log-based query suggestions.
“Entities,” as used herein are items having known information (e.g., categories, attributes, and attribute values) associated therewith that is organized and stored in the entity store 206. In some instances, entities have associated categories, but in another instances they do not. “Categories,” “attributes” and “attribute values,” as these terms are utilized herein, represent various hierarchical characteristics of entities and are useful in filtering or formulating received search queries (or portions thereof) for providing a user with as accurate an answer to a particular query as possible. While an entity is described herein as a particular product in one embodiment, in other embodiments, entities may also include other items that are not considered “products.” For instance, for exemplary purposes only, entities may include a particular product, a business, an organization, a person, or the like. Further, an entity may be a specific location that has associated categories, attributes, and attribute values from which the user may select to formulate the search query. The definition of an entity, as used herein, is not meant to be limiting, but may apply to many things that are potentially capable of having associated hierarchical characteristics.
By way of example, suppose a user enters the character sequence “role” into the search query input area. As a result, “rolex” may be identified as a potential query formulation suggestion by either identifying it as a query-log-based query suggestion or as a semantic-pattern-based query suggestion. Here, “rolex” may be identified as a brand of watches, or may be a popular query, and may also correspond to one or more entities found in the entity index 206. The entity “rolex” may be associated with the category “watches,” and such category may be returned to the user as a suggested category that may then be utilized to further formulate the user's search. Similarly, if a user enters the alphanumeric characters “facebo” into the search query input area, it may lead to the identification of the query formulation suggestion “facebook,” which may correspond to one or more entities in the entity store 206. Further, the entities may include associated categories, such as “Web,” such that an entered search query for “facebo” may return “Web” as a suggested category. This provides the user with an indication as to how the portion of the entered search query has been interpreted.
In embodiments, a received search query may be identical or substantially similar to an identified category. For example, the search query “digital cameras” may result in a plurality of entities having a category “digital cameras.” While this would not in and of itself necessarily help in refining the user's search, armed with the knowledge that the category matches the intent of the user's search, various lower level hierarchical characteristics, e.g., attributes and attribute values, associated with the identified entity may be suggested to the user as possible query formulation suggestions. As another example, suppose the user enters the alphanumeric characters “digital slr” as a portion of a search query into the search query input area. The system of embodiments of the present invention may access the entity store 206 to determine one or more entities corresponding to the query “digital slr.” Since there are several models of cameras that are SLR, there are likely several entities, if not hundreds of entities, that may be identified. The category in this instance may be “digital cameras.” The previous description of categories may apply to even those situations where the query is not substantially similar to a stored category.
In embodiments, the entity store 206 is configured to be searchable for one or more of the items stored in association therewith. It will be understood and appreciated by those of ordinary skill in the art that the information stored in association with the entity store 206 may be configurable and may include any information relevant to entities, categorizations of entities, characteristics of entities, and the like. The content and volume of such information are not intended to limit the scope of embodiments of the present invention in any way. Further, though illustrated as a single, independent component, the entity store 206 may, in fact, be a plurality of storage devices, for instance, a database cluster, portions of which may reside on the semantic suggestion generator 208, another external computing device (not shown), and/or any combination thereof.
As previously stated, the entity store 206 is configured to store attributes for entities as they pertain to particular categories. Attributes are characteristics of the entities belonging to a particular category that may be suggested to a user to further describe or refine the user's intent upon entering a search query. For example, attributes for the entity “Canon PowerShot SX30” may be associated with the category “digital cameras.” The entity “Canon PowerShot SX30,” by way of example only, may be associated with the attributes “megapixels,” “color,” “zoom,” “lens,” “viewfinder type,” “image stabilization,” “brand,” “type,” “model,” and the like. These attributes may be used to further refine a search query for “Canon cameras,” for example, to aid in discovering the user's true intent.
In addition to entities, categories and attributes for entities, the entity store 206 also is configured to store attribute values as they pertain to particular entities and associated attributes. For instance, for the attribute “color” associated with the entity “Canon PowerShot SX30,” attribute values may include, for instance, “black,” “blue,” “red,” “silver,” etc. In general, values to suggest are identified by considering the distribution of value for all entities that match the current query. For example, for “canon cameras,” there may be over 100 camera entities with “brand=canon.” Among these, some are tagged with “color=black.” Others may be tagged with “color=silver.” The distribution of these values along with the score and popularity of the entities is used to determine the set and ranking of values presented to the user. Attribute values may be suggested to the user as query formulation suggestions to aid in further describing or refining the user's intent upon receiving a search query. This effectively blends browsing and searching experiences of web scale databases. As mentioned, attributes and attribute values are dependent upon the particular entity with which they are associated, and also are dependent upon the current state of the search query. For instance, if the search query already has a color specified, the attribute “color” may not even be presented to the user. Similarly, if the user has specified in the search query or has already selected an attribute value of “Canon,” the system may not present the attribute of “brand” to the user.
While described in the context of search queries and assisting the user to define a search query, embodiments of the present invention may also be used in the context of related search, related categories, related concepts, etc. For instance, in one embodiment, alternative categories may be displayed for the user in addition to the best category associated with the entities that correspond to the search query. These alternative categories can be selected by the user if the user is not satisfied with the category chosen by the system. This functionality is further described below.
With continued reference to
The user device 202 may be any type of computing device owned and/or operated by an end user that can access the network 210. For instance, the user device 202 may be a desktop computer, a laptop computer, a tablet computer, a mobile device, or any other device having network access. Generally, an end user may employ the user device 202 to, among other things, access electronic documents by submitting a search query to the search engine 204. For instance, the end user may employ a web browser on the user device 202 to access and view electronic documents stored in the system 200. According to embodiments of the present invention, the user additionally may be able to interact with the user device 202 by selecting an attribute and an attribute value, which further refines an input search query, as more fully described below. While the user device 202 has been described as communicating by way of the network 210, embodiments of the present invention may also be used locally on a device, such as to improve desktop search, email search, or application search on Smartphone, and as such a network, such as the network 210, may not be required.
The search engine 204 may comprise multiple components that are responsible for receiving an entered search query and identifying one or more documents that are relevant to that particular search query. In embodiments of the present invention, multiple sets of relevant documents may be presented to the user over the course of time from when the search query is first entered, to when a satisfactory response is received by the user through search query formulations utilizing attributes, attribute values, and the like. As such, the search engine 204 and the semantic suggestion generator 208, in one embodiment, are generally in communication with one in another while a particular search query is being developed or completed. In an alternative embodiment, the search engine 204 and the semantic suggestion generator 208 are not associated with each other. For instance, at each semantic boundary (e.g., completion of a term in the search query, addition of an attribute value to the search query), the search results are updated in one embodiment. The search engine 204 may utilize one or more search indexes that have indexed documents from the web so that the most relevant search results can be identified in an efficient manner. In addition to web documents, the search engine may provide results from semi-structured databases based on the semantic constraints specified in the query.
As mentioned, in one embodiment, search results are updated when a semantic boundary is reached, such as when a term in the search box has been completed. This may occur when a new term has been added to the search query by either user input or by a user selection of a query formulation suggestion, which causes the search query to be modified according to the selected query formulation suggestion.
As shown in
Additional components of the semantic suggestion generator 208 (not shown) may include, without limitation, a processing unit, internal system memory, and a suitable system bus for coupling various system components, including one or more data stores for storing information (e.g., files and metadata associated therewith). The semantic suggestion generator 208 typically includes, or has access to, a variety of computer-readable media. By way of example, and not limitation, computer-readable media may include computer-storage media and communication media. In general, communication media enables each computing device to exchange data via a network, e.g., network 210. More specifically, communication media may embody computer-readable instructions, data structures, program modules, or other data in a modulated data signal, such as a carrier wave or other transport mechanism, and may include any information-delivery media. As used herein, the term “modulated data signal” refers to a signal that has one or more of its attributes set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above also may be included within the scope of computer-readable media.
It will be understood by those of ordinary skill in the art that computing system environment 200 is merely exemplary. While the semantic suggestion generator 208 is illustrated as a single unit, one skilled in the art will appreciate that the semantic suggestion generator 208 is scalable. For example, the semantic suggestion generator 208 may in actuality include a plurality of computing devices in communication with one another. Moreover, the entity store 206, or portions thereof, may be included within, for instance, the semantic suggestion generator 208, a Website source, or a third-party service as a computer-storage medium. The single unit depictions are meant for clarity, not to limit the scope of embodiments in any form.
The receiving component 212 is configured to receive search queries, or portions thereof, input by users (e.g., users of user device 202). Such information may be received in real-time upon entry by a user. Search queries may also be received by the receiving component 212 from one or more query logs having historical query information included therein for one or both of a particular user and a group of users as a whole. Received search queries, or portions thereof, may be semantically analyzed, as further described herein, to identify one or more query formulation suggestions to present to the user as suggestions for formulating the current search query. Semantically analyzing at least a portion of a search query comprises hypothesizing the user intended task or category and extracting specified attribute values by tagging words in the query with their attributes. For instance, the query “red canon digital camera” may be identified as a category=digital camera, brand=canon, and color=red. Query formulation suggestions may be generated, in one instance, using templates associated with queries, attributes, categories, attribute values, etc. Thus, the receiving component 212 is additionally configured to receive information from the entity store 206. The receiving component 212 is further configured to receive indications of user selections of query formulation suggestions. Such indications may be received, by way of example only, from user device 202.
The semantic pattern identifying component 214 is configured to identify semantic patterns associated with queries, categories, attributes and attribute values. In embodiments, the semantic pattern identifying component 214 identifies key terms/phrases commonly associated with categories corresponding to particular entities. As mentioned, semantic patterns are similar to templates, in that at least a portion of a semantic query pattern identifies a particular category or attribute that is filled in based on the received search query. Templates may be stored in a semantic grammar index, for instance, as shown herein in
In embodiments, semantic query patterns may be identified algorithmically from information stored in query logs and/or the entity store 206. Semantic query patterns, that is, linguistic patterns contained in received queries, will vary among languages and are determined based upon knowledge of how particular entities are commonly phrased in the language in which a search query is received. Alternatively or additionally, for instance in situations where standard language-specific linguistic patterns are unable to be identified or are clearly not followed, semantic query patterns of interest may be identified manually via editorial intervention.
In embodiments, weights or probabilities (that is, the likelihood of a particular semantic query pattern being a pattern that represents the true intent of a user) also may be determined by the semantic pattern identifying component 214. In embodiments, such weights may be determined based upon one or both of availability and interest. In terms of availability, the semantic pattern identifying component 214 may examine the data contained in the entity store 206 and estimate how frequently particular items or entities are likely to occur. For instance, in a commerce setting, the semantic pattern identifying component 214 may examine the data on product availability that is contained in the entity store 206 and estimate how frequently each entity is likely to be offered or available. The more frequently a particular entity will be offered or available, the higher the probability or weight assigned to that entity will be.
In terms of interest, the semantic pattern identifying component 214 may examine input user queries and/or user-behavior with respect to the entities stored in association with the entity store 206, and generate a weight that represents relative user interest in a particular entity. For instance, in the commerce setting, the semantic pattern identifying component 214 may examine user queries and/or behavior with respect to a particular product and estimate the relative interest of users in that product. Products with higher user interest will then be assigned a higher probability or weight. In embodiments wherein both availability and interest are utilized to determine weights or probabilities of given entities/items, a single weight representing some combination of the availability weight and the user interest weight may be determined.
In embodiments, semantic attribute value patterns may be identified algorithmically from information stored in query logs and/or the entity store 206. Semantic attribute value patterns, that is, linguistic patterns contained in attribute values, will vary among languages and are determined based upon knowledge of how particular entities are commonly phrased in the language in which an attribute value is identified. Alternatively or additionally, for instance in situations where standard language-specific linguistic patterns are unable to be identified or clearly are not followed, semantic attribute value patterns of interest may be identified manually via editorial intervention.
The constraint index 216 is configured to extract entities from the entity store 206 with their respective categories, attributes, attribute values and corresponding weights and to determine unreasonable semantic patterns for each entity—that is, to determine semantic patterns that are illogical or are highly unlikely. Thus, if a particular category is rarely in the entity store 206 associated with an entity having a particular attribute, that category/attribute pair would be determined to be illogical for the particular entity. More commonly, constraints among attributes within a particular category may be applied such that the term “Microsoft PlayStation” is not suggested because the brand “Microsoft” and the product “PlayStation” are incompatible. The constraint index 216 may be utilized to filter query formulation suggestions, which will be discussed further herein. In some instances, the constraint index 216 also filters semantic patterns identified by the semantic pattern identifying component 214 if an attribute is inconsistent with currently specified constraints. This functionality is more fully described below. In embodiments, the constraint index 216 is further configured to build a compressed binary representation of the extracted data.
In embodiments, the constraint index 216 is configured to build a compressed binary entity index that allows the semantic suggestion generator 208 to efficiently determine the top attributes and attribute values matching a given set of constraints in a particular category and an optional prefix, along with their weights. In this way, hypotheses that cannot be true can be quickly eliminated leading to increased efficiency in arriving at suggestions to aid in determining user intent. For instance, if the user is searching a directory that is organized alphabetically and the user inputs the character “a,” the compressed binary entity index can quickly filter the directory and eliminate all items not beginning with the character “a.”
The category identifying component 218 is configured to identify the top category associated with a query. In embodiments, the category identifying component 218 is further configured to identify one or more categories that are related to the query, but that may not have been initially selected as the category thought to be most relevant to the search query. In embodiment, categories are stored in the entity store 206 in relation to their associated entities. In some embodiments, an entity may not have an associated category, and as such a category is not displayed on the user interface.
Similarly, the attribute identifying component 220 is configured to identify at least one attribute associated with an entity. Attributes, as used herein, classify different attributes associated with an entity, and are specific to the entity to which they correspond. For instance, one entity may be associated with a particular attribute, but that attribute may not be associated with a different entity. In embodiments the attribute identifying component 220 is further configured to identify one or more attributes that are related to the entity, but that may have not been identified as one of the most relevant attributes. Further, the attribute value identifying component 222 is configured to identify at least one attribute value associated with an entity. Attribute values, as used herein, are values that correspond directly to the attribute and entity to which they correspond. For example, if a specific type of watch does not come in the color purple, “purple” is not displayed to the user as an attribute value corresponding to the attribute “color.” For example, one entity may be associated with a particular attribute, but that attribute may not be associated with a different entity. In embodiments the attribute value identifying component 222 is further configured to identify one or more attribute values that are related to the entity but may not have been identified as one of the most relevant attributes.
The query formulation suggestion generating component 224, given a search query (or portion thereof), is configured to construct a lattice of possible interpretations of the input query (or query portion). The lattice is dynamically composed utilizing the semantic patterns identified by the semantic pattern identifying component 214 and applying a search algorithm (for instance, the A* search algorithm well known to those of ordinary skill in the art) to find the search query completion path or paths that represent the most likely intent of the user while preserving the constraints imposed by the constraint index 216. These types of algorithms can also find extensions of a partial inputted search query. Those completion paths (refined queries) identified as most likely intended by the user are those that are identified as having the highest overall weights, that is, the highest probabilities.
The transmitting component 226 is configured to transmit query formulation suggestions, which may take the form of query-log-based query suggestions, semantic-pattern-based query suggestions, entities, categories, attributes, attribute values, or the like for presentation. For instance, the query “digital ca” may result in a query-log-based query suggestion of “digital camera.” Similarly, the query “canon in ca.” may result in a category completion (e.g., query formulation suggestion) of “canon in cameras.” Further, the query “cameras by br” may result in the attribute completion (e.g., query formulation suggestion) of “cameras by brand.” The query “cameras by can” may result in an attribute value completion (e.g., query formulation suggestion) of “cameras by canon.”
The semantic analyzing component 228 is configured to semantically analyze a received search query, or a portion thereof in order to identify the most relevant query formulation suggestions based on the received search query. In embodiments, algorithms are used to semantically analyze search queries. In one embodiment, as previously described, semantic-pattern-based query suggestions are derived using semantic query patterns and the entity store. For instance, semantic-pattern-based query suggestions may be compared to known entities in the entity store prior to being transmitted for presentation. For example, for the partial query “canon ca,” the system searches through query records, query/attribute templates, category, etc. to find possible completions of the query that are also consistent with the known information in the entity store 206. Suppose that “<brand>camera” is in the “digital camera” category and “<brand>car stereos” is in the “car stereos” category as query templates. Possible semantic-pattern-based query suggestions may include, for exemplary purposes only, “canon camera” and “canon car stereos.” However, when consulting with the entity store 206, it is determined that there is no such entity associated with the category “car stereos” with the brand “canon.” This indicates that “canon car stereos” is an invalid query formulation suggestion, and thus is not provided as a suggestion.
In some instances, multiple query templates are stored for different categories that match the same query. For example, the partial query “can” may be completed to “canon” using the template “<brand>” in both the “digital camera” and “printers” category. In other words, there are entities in both categories with “canon” as the brand. In such scenarios, we show category disambiguation for the completion suggestion “canon,” as further described herein. It is possible that there are other completions for “can” that are also ambiguous, such as “candy” as a type of food or the name of a movie. Thus, within a single suggestion list, there may be multiple completion suggestions that are category ambiguous.
Once potential query completion suggestions are identified, if a user selects a completion or if the system has a high enough confidence for a particular completion, the system may further identify query formulation suggestions for that particular completion about which the system has a high degree of confidence. For example, if the system knows that “canon camer” completes to “canon camera” with a high probability, query formulation suggestions such as “canon camera by resolution,” “canon camera by color,” etc., may be presented to the user. Even if the user enters “canon camera,” the system may still first try to complete the query before suggestion query formulation suggestions.
Utilizing the components 212, 214, 216, 218, 220, 222, 224, 226, and 228, the semantic suggestion generator 208 is configured to construct a trie from one or more query logs, each query being annotated with its top weighted interpretations. Then, given a partial query, the semantic suggestion generator 208 is able to locate the top weighted completion paths from the trie and transmit them for presentation as query formulation suggestions. In instances wherein not enough suggestions are determined to qualify as “top” suggestions, for instance where a minimum threshold number of query formulation suggestions is desired, additional query formulation suggestions may be identified by the query formulation suggestion generating component 224 (even though they may not meet a minimum desired threshold weight, for instance). Any semantically duplicate suggestions may be eliminated. Because a trie structure is utilized, if a category/attribute/attribute value suggestion is the semantic prefix of one or more remaining suggestions, an attribute extension suggestion may be inserted below such category/attribute/attribute value suggestion upon presentation. In embodiments, the semantic suggestion generator 208 may render query trie suggestions with multiple interpretations as an annotated query with disambiguation information. These embodiments are more fully described below with reference to the screen displays of
Turning now to
A suggestion area 315, as shown in
Referring to
In one embodiment, in addition to user a mouse or keyboard to select query formulation suggestions, the user may also select/filter these suggestions by typing or otherwise entering the name/template of the query formulation suggestion directly. For instance, the user may type “canon camera by res” or “canon camera 10 megap.” In such instances, query completion suggestions may be presented that attempt to complete the attribute or value. Depending on the scenario, there may be multiple query completion suggestions such a scenario. For instance, a partial query of “canon camera b” may result in the query completion suggestions of, for example, “canon camera black,” “canon camera blue,” etc. There may also be additional query completion suggestions from previously seen queries (e.g., popular queries), such as “canon camera bag.”
In one embodiment, while the user may initially enter the search query “digital camera” and eventually add more attribute values to the search query using the methods described herein, the system may reformulate or paraphrase the search query for the user as more information is added. So instead of “digital camera Nikon,” the system may reformulate it to read “Nikon digital camera,” which likely appeals more to the user as it is in a natural language format. Not only may the query be reformulated in the search box, but it may also be in reformulated format on the search results page.
Turning now to
The query formulation suggestions may be generated based on the identified entities. The query formulation suggestions may include, for example, categories, attributes, and attribute values which assist the user to formulate the search query. The query formulation suggestions are transmitted for presentation at step 2714. As mentioned, query formulation suggestions may include categories, attributes, and attribute values. In some embodiments, there may not be any categories associated with an entity. In that case, no categories are presented to the user. But if any attributes or attributes values are known in relation to that entity, those may be presented to the user. The identified attribute suggestions may be displayed in a suggestion area, such as the suggestion area illustrated in various figures herein. In one embodiment, query-log-based query suggestions are also displayed in the suggestion area in addition to the attribute suggestions to provide the user with a blended experience that encompasses both query-log-based query suggestions (e.g., popular queries that have been submitted to the search engine in the past) and query formulation suggestions. The query formulation suggestions are configured to be selectable by a user. For instance, when the user selects a particular attribute value, such as “blue,” the term “blue” is added to the search query to further define the user's intent.
Referring to
Query formulation suggestions associated with the identified entities are identified, and may include category suggestions, attribute suggestions, attribute value suggestions, etc. Categories may include product types (e.g., digital camera, video camera, watches, party supplies, smart phones) or may even include non-products, such as locations, ideas, or the like. Generally, attributes are groups of different aspects corresponding to a particular entity that are found or known to be relevant to that entity. At step 2814, the query formulation suggestions are transmitted for presentation in, for example, a suggestion area, and are configured to be selected by a user to further formulate the search query. In one embodiment, the suggestion area is located on the user interface beneath the search box, but in other embodiments, is located in another location on the user interface.
A user selection is received of one of the query formulation suggestions at step 2816. This user selection may include a hard selection (e.g., pressing the enter key on the keyboard, mouse clicking the attribute) or a soft selection (e.g., hovering over the attribute). In one embodiment, the user selection of an attribute is a hover of the mouse over the attribute, which causes attribute value suggestions to be displayed, as discussed below. Once a user selection is made, the query is modified at step 2818 based on the selected query formulation suggestion.
Attribute values are simply values that correspond to the selected attribute. For example, an attribute of color would potentially have values including, for example, blue, green, orange, red, etc. Attribute values are selectable by the user and help to further define and formulate the user's search query. It should be noted that both attribute suggestions and attribute value suggestions are dynamic lists and change based on the current form of the user's search query, category, etc. For instance, attribute values for the attribute “color” are different based on whether the category is a laptop or a car. Laptops made by Dell, for example, likely do not come in the same colors as a particular vehicle Likewise, laptops made by one manufactures may not come in the same colors as laptops made by another manufacture. Additionally, once a user has further defined the search query, the attribute associated with the additional term added to the search query may not be visible in the list of attribute suggestions, since the user has already made a decision based on that particular attribute.
With reference now to
The text-parser may be utilized in parsing input user queries or portions thereof, in accordance with embodiments herein above described. In one embodiment in this regard, at least a portion of an input user search query is received, the text-parser is applied to determine at least one likely attribute for expansion of the input user query, the likelihood being based upon the respective weights, and the at least one likely attribute is transmitted for presentation as an attribute suggestion. In embodiments, applying the text-parser to determine at least one likely attribute for expansion of the input user query may comprise applying the text-parser to determine a plurality of likely attributes for expansion of the input user query, each of the plurality of likely attributes having a calculated weight associated therewith. In embodiments, transmitting the at least one likely attribute for presentation as an attribute suggestion may comprise transmitting at least a portion of the plurality of likely attributes for presentation as attribute suggestions, the attribute suggestions representing the plurality of likely attributes or portion thereof that are transmitted for presentation being arranged in an order representing their respective calculated weights.
As can be seen, embodiments of the present invention provide for systems, methods and computer-storage media having computer-usable instructions embodied thereon, for building, linking, and exposing semantic knowledge of a search system or systems (e.g., query logs, facets, relationships of entities from structured and unstructured data, contextual signals, and the like) to assist users in query formulation and intent disambiguation, as well as query formulation. Embodiments of the invention may be scaled to tail and never-seen queries and can provide better user intent signals to the system for result matching and lead to better and more relevant search results.
Turning now to
Turning now to
The present invention has been described in relation to particular embodiments, which are intended in all respects to be illustrative rather than restrictive. Alternative embodiments will become apparent to those of ordinary skill in the art to which the present invention pertains without departing from its scope.
From the foregoing, it will be seen that this invention is one well adapted to attain all the ends and objects set forth above, together with other advantages which are obvious and inherent to the system and method. It will be understood that certain features and subcombinations are of utility and may be employed without reference to other features and subcombinations. This is contemplated by and is within the scope of the claims.
This application claims the benefit of Provisional Patent Application No. 61/476,154 filed on Apr. 15, 2011 and entitled “Interactive Semantic Query Suggestion for Content Search. This application is further related by subject matter to U.S. patent application Ser. No. ______, entitled Interactive Semantic Query Suggestion for Content Search,” filed on even date herewith (attorney docket number MS 332652.02/MFCP.162846). Both Provisional Patent Application No. 61/476,154 and U.S. patent application Ser. No. ______ (attorney docket number MS 332652.02/MFCP.162846) are hereby incorporated by reference as is set forth in their entirety herein.
| Number | Date | Country | |
|---|---|---|---|
| 61476154 | Apr 2011 | US |
