The disclosed embodiments relate generally to fact repositories. More particularly, the disclosed embodiments relate to determining whether different objects in a repository are associated with the same entity.
When data are gathered from diverse sources, it is often difficult to determine whether data from different sources pertain to the same entity. For example, consider the problem of extracting data from web pages and other electronic documents on the Internet in order to build a repository of objects containing facts about entities. Generally, it is possible to analyze a web page and identify the name of the entity that the page describes. For example, one can determine that a web page describes the entity named “George Bush.” Therefore, one technique for building the fact repository is to create an object for each (name, web page) tuple and associate all of the facts on the given web page with that object.
Since the technique described above treats each object formed from a (name, web page) tuple as unique, it can result in many different objects associated with the same entity. There might be 7,000,000 web pages references for “George Bush,” 5,000,000 references for “Bill Clinton,” and an additional 500,000 references for “William Jefferson Clinton,” and each web page results in a separate object. However, some objects with the same name might be associated with different entities. For example, two objects named “George Bush” can be associated with different entities if one object references the 41st President of the United States while the other references the 43th President. Likewise, two objects named “Bill Clinton” can be associated with different entities if one object describes the 42nd President while the other describes a book about the Clinton presidency. Two objects with different names might also describe the same entity. Additional complications arise because even objects about the same entity are likely to contain different subsets of facts about the entity, and objects will sometimes contain erroneous facts due to errors in the source documents.
Ideally, the fact repository should contain exactly one extracted object for each unique entity. However, the large number of web pages and resulting extracted objects makes it impractical for human users to review and analyze the objects in the repository
The above and other needs are met by methods, systems, and computer program products that merge objects associated with the same entity. Embodiments of the method comprise identifying a plurality of merge candidate objects, each merge candidate object including one or more facts describing an entity with which the object is associated and having at least one fact with a same attribute, and grouping the plurality of merge candidate objects responsive to the values of the fact having the same attribute. The method further identifies similarities between objects in each group and generates one or more graphs describing the identified similarities among the objects of all of the groups. The method analyzes the one or more graphs describing the similarities among the objects to identify two or more objects associated with the same entity. The method also merges the two or more objects associated with the same entity to produce a merged object and stores the merged object in a repository.
Embodiments of the system and computer program product comprise a grouping module for identifying a plurality of merge candidate objects, each merge candidate object including one or more facts describing an entity with which the object is associated and at least one fact with a same attribute. The grouping module also groups the plurality of merge candidate objects responsive to values of the fact having the same attribute. The system and computer program product further comprise an object comparison module for identifying similarities between objects in each group and a graph generation module for generating one or more graphs describing identified similarities among the objects of all of the groups. Finally, a merging module analyzes the one or more graphs describing the similarities among the objects to identify two or more objects associated with the same entity, merges the two or more objects associated with the same entity to produce a merged object, and stores the merged object in a repository.
a)-2(d) are block diagrams illustrating a data structure for facts within a repository of
e) is a block diagram illustrating an alternate data structure for facts and objects in accordance with some embodiments.
Embodiments are now described with reference to the figures where like reference numbers indicate identical or functionally similar elements.
Document hosts 102 store documents and provide access to documents. A document is comprised of any machine-readable data including any combination of text, graphics, multimedia content, etc. A document may be encoded in a markup language, such as Hypertext Markup Language (HTML), i.e., a web page, in an interpreted language (e.g., JavaScript) or in any other computer readable or executable format. A document can include one or more hyperlinks to other documents. A typical document will include one or more facts within its content. The facts describe entities, such as a real-world or fictional people, places, or things.
A document stored in a document host 102 may be located and/or identified by a Uniform Resource Locator (URL), or Web address, or any other appropriate form of identification and/or location. A document host 102 is implemented by a computer system, and typically includes a server adapted to communicate over the network 104 via networking protocols (e.g., TCP/IP), as well as application and presentation protocols (e.g., HTTP, HTML, SOAP, D-HTML, JAVA®). The documents stored by a host 102 are typically held in a file directory, a database, or other data repository. A host 102 can be implemented in any computing device (e.g., from a PDA or personal computer, a workstation, mini-computer, or mainframe, to a cluster or grid of computers), as well as in any processor architecture or operating system.
Janitors 110 operate to process facts extracted by the importer 108. This processing can include but is not limited to, data cleansing, object merging, and fact induction. In one embodiment, there are a number of different janitors 110 that perform different types of data management operations on the facts. For example, one janitor 110 may traverse some set of facts in the repository 115 to find duplicate facts (that is, facts that convey the same factual information) and merge them. Another janitor 110 may also normalize facts into standard formats. Another janitor 110 may also remove unwanted facts from the repository 115, such as facts related to pornographic content. Other types of janitors 110 may be implemented, depending on the types of data management functions desired, such as translation, compression, spelling or grammar correction, and the like.
Various janitors 110 act on facts to normalize attribute names, and values and delete duplicate and near-duplicate facts so an object does not have redundant information. For example, we might find on one page that Britney Spears' birthday is “Dec. 2, 1981” while on another page that her date of birth is “Dec. 2, 1981.” Birthday and Date of Birth might both be rewritten as “Birthdate” by one janitor and then another janitor might notice that Dec. 2, 1981 and Dec. 2, 1981 are different forms of the same date. It would choose the preferred form, remove the other fact and combine the source lists for the two facts. As a result, one source page for this fact will contain an exact match of the fact while another source page will contain text that is considered synonymous with the fact.
The build engine 112 builds and manages the repository 115. The service engine 114 is an interface for querying the repository 115. The service engine 114's main function is to process queries, score matching objects, and return them to the caller but it is also used by the janitor 110.
The repository 115 stores factual information about entities. The information is extracted from a plurality of documents that are located on document hosts 102. A document from which a particular fact may be extracted is a source document (or “source”) of that particular fact. In other words, a source of a fact includes that fact (or a synonymous fact) within its contents.
The repository 115 contains one or more facts. In one embodiment, the facts are logically organized into “objects,” and each object contains a collection of facts associated with a single entity (i.e., real-world or fictional person, place, or thing). Each fact is associated with exactly one object. One implementation for this association includes in each fact an object ID that uniquely identifies the associated object. In this manner, any number of facts may be associated with an individual object, by including the object ID for that object in the facts. In one embodiment, objects themselves are not physically stored in the repository 115, but rather are defined by the set or group of facts with the same associated object ID, as described below. Further details about facts in the repository 115 are described below, in relation to
Some embodiments operate on the facts and/or objects in different orders than described above. For example, in one embodiment the importer 108 provides facts directly to the build engine 112 and/or repository 115. The janitors 110, in turn, operate on the facts and/or objects in the repository 115. It should also be appreciated that in practice at least some of the components of the data processing system 106 will be distributed over multiple computers, communicating over a network. For example, the repository 115 may be deployed over multiple servers. As another example, the janitors 110 may be located on any number of different computers. For convenience of explanation, however, the components of the data processing system 106 are discussed as though they were implemented on a single computer.
In another embodiment, some or all of document hosts 102 are located on the data processing system 106 instead of being coupled to the data processing system 106 by a network. For example, the importer 108 may import facts from a database that is a part of or associated with the data processing system 106.
a) shows an example format of a data structure for facts within the repository 115, according to some embodiments. As described above, the repository 115 includes facts 204 describing entities such as real-world and fictional people, places, and things. Each fact 204 includes a unique identifier for that fact, such as a fact ID 210. Each fact 204 includes at least an attribute 212 and a value 214. For example, a fact associated with the entity George Washington may include an attribute of “date of birth” and a value of “Feb. 22, 1732.” In one embodiment, all facts are stored as alphanumeric characters since they are extracted from web pages. In another embodiment, facts also can store binary data values. Other embodiments, however, may store fact values as mixed types, or in encoded formats.
As described above, each fact is associated with an object ID 209 that identifies the object with which the fact is associated. Thus, each fact that describes the same entity (such as George Washington), will have the same object ID 209. In one embodiment, the objects are logical concepts that exist as a collection of facts having the same object ID. In another embodiment, objects are stored as units of data in memory, and include references (for example, pointers or IDs) to the facts associated with the object. The logical data structure of a fact can take various forms; in general, a fact is represented by a tuple that includes a fact ID, an attribute, a value, and an object ID. The storage implementation of a fact can be in any underlying physical data structure.
b) shows an example of facts having respective fact IDs of 10, 20, and 30 in the repository 115. Facts 10 and 20 are associated with an object identified by object ID “1.” Fact 10 has an attribute of “Name” and a value of “China.” Fact 20 has an attribute of “Category” and a value of “Country.” Thus, the object identified by object ID “1” has a name fact 205 with a value of “China” and a category fact 206 with a value of “Country.” Fact 30208 has an attribute of “Property” and a value of “Bill Clinton was the 42nd President of the United States from 1993 to 2001.” Thus, the object identified by object ID “2” has a property fact with a fact ID of 30 and a value of “Bill Clinton was the 42nd President of the United States from 1993 to 2001.” In the illustrated embodiment, each fact has one attribute and one value. The number of facts associated with an object is not limited; thus while only two facts are shown for the “China” object, in practice there may be dozens, even hundreds of facts associated with a given object. Also, the value fields of a fact need not be limited in size or content. For example, a fact about the economy of “China” with an attribute of “Economy” would have a value including several paragraphs of text, numbers, and perhaps even tables of figures. This content can be formatted, for example, in a markup language. For example, a fact having an attribute “original html” might have a value of the original html text taken from the source web page.
Also, while the illustration of
c) shows an example object reference table 210 that is used in some embodiments. Not all embodiments include an object reference table. The object reference table 210 functions to efficiently maintain the associations between object IDs and fact IDs. In the absence of an object reference table 210, it is also possible to find all facts for a given object ID by querying the repository 115 to find all facts with a particular object ID. While
d) shows an example of a data structure for facts within the repository 115, according to some embodiments, showing an extended format of facts. In this example, the fields include an object reference link 216 to another object. The object reference link 216 can be an object ID of another object in the repository 115, or a reference to the location (e.g., table row) for the object in the object reference table 210. The object reference link 216 allows facts to have as values other objects. For example, for an object associated with the entity “United States,” there may be a fact with the attribute of “president” and the value of “George W. Bush,” with “George W. Bush” being an object having its own facts in the repository 115. In some embodiments, the value field 214 stores the name of the linked object and the link 216 stores the object identifier of the linked object. Thus, this “president” fact would include the value 214 of “George W. Bush”, and an object reference link 216 that contains the object ID for the “George W. Bush” object. In some other embodiments, facts 204 do not include a link field 216 because the value 214 of a fact 204 may store a link to another object.
Each fact 204 also may include one or more metrics 218. A metric provides an indication of the quality of the fact. In some embodiments, the metrics include a confidence level and an importance level. The confidence level indicates the likelihood that the fact is correct. The importance level indicates the relevance of the fact to the object, compared to other facts for the same object. The importance level may optionally be viewed as a measure of how vital a fact is to an understanding of the entity associated with the object.
Each fact 204 includes a list of one or more sources 220 that include the fact and from which the fact was extracted. Each source may be identified by a Uniform Resource Locator (URL), or Web address, or any other appropriate form of identification and/or location, such as a unique document identifier.
The facts illustrated in
Some embodiments include one or more specialized facts, such as a name fact 207 and a property fact 208. A name fact 207 is a fact that conveys a name for the entity associated with the object in which the fact is included. A name fact 207 includes an attribute 224 of “name” and a value, which is the name of the associated entity. For example, for an object associated with country Spain, a name fact would have the value “Spain.” A name fact 207, being a special instance of a general fact 204, includes the same fields as any other fact 204; it has an attribute, a value, a fact ID, metrics, sources, etc. The attribute 224 of a name fact 207 indicates that the fact is a name fact, and the value is the actual name. The name may be a string of characters. An object may have one or more associated name facts, as many entities can have more than one name. For example, an object associated with Spain may have associated name facts conveying the country's common name “Spain” and the official name “Kingdom of Spain.” As another example, an object associated with the U.S. Patent and Trademark Office may have associated name facts conveying the agency's acronyms “PTO” and “USPTO” as well as the official name “United States Patent and Trademark Office.” If an object does have more than one associated name fact, one of the name facts may be designated as a primary name and other name facts may be designated as secondary names, either implicitly or explicitly.
A property fact 208 is a fact that conveys a statement about the entity associated with the object. Property facts are generally used for summary information about an object. A property fact 208, being a special instance of a general fact 204, also includes the same fields (such as attribute, value, fact ID, etc.) as other facts 204. The attribute field 226 of a property fact 208 indicates that the fact is a property fact (e.g., attribute is “property”) and the value is a string of text that conveys the statement of interest. For example, for the object associated with Bill Clinton, the value of a property fact may be the text string “Bill Clinton was the 42nd President of the United States from 1993 to 2001.” Some objects may have one or more associated property facts while other objects may have no associated property facts. It should be appreciated that the data structures shown in
As described previously, a collection of facts is associated with an object ID of an object. An object may become a null or empty object when facts are disassociated from the object. A null object can arise in a number of different ways. One type of null object is an object that has had all of its facts (including name facts) removed, leaving no facts associated with its object ID. Another type of null object is an object that has all of its associated facts other than name facts removed, leaving only its name fact(s). Alternatively, the object may be a null object only if all of its associated name facts are removed. A null object represents an entity or concept for which the data processing system 106 has no factual information and, as far as the data processing system 106 is concerned, does not exist. In some embodiments, facts of a null object may be left in the repository 115, but have their object ID values cleared (or have their importance set to a negative value). However, the facts of the null object are treated as if they were removed from the repository 115. In some other embodiments, facts of null objects are physically removed from the repository 115.
e) is a block diagram illustrating an alternate data structure 290 for facts and objects in accordance with embodiments of the invention. In this data structure, an object 290 contains an object ID 292 and references or points to facts 294. Each fact includes a fact ID 295, an attribute 297, and a value 299. In this embodiment, an object 290 actually exists in memory 107.
As described above, an object may explicitly exist in the repository 115, or it may exist merely as a collection of facts with a common object ID. Reference is made to particular objects for the purposes of illustration; one of skill in the art will recognized that the systems and methods described herein are applicable to a variety of implementations and that such references are not limiting. When reference is made to a fact being associated with an object, it should be understood that in at least one embodiment a fact is associated with an object by sharing a common object ID with other facts. For example, a fact could be associated with an object of a given type by sharing a common object ID at least with a type fact indicating the given type (or as another example, with a category fact indicating a particular category of object). Furthermore, in various embodiments, facts and objects can be stored in a variety of structures, such as fact and/or object repositories. When reference is made herein to the repository 115, it should be understood that various embodiments may store facts and/or objects in a variety of data structures.
An embodiment of the object merge engine 300 is implemented as a janitor 110. However, the object merge engine 300 can also be implemented by other entities shown in
A grouping module 310 groups the merge candidate objects based on the values of their facts. In one embodiment, all of the merge candidate objects include at least one fact having a same attribute. For example, each object includes a name fact having a “name” attribute and a corresponding value describing the name of the entity with which the object is associated. One embodiment creates a group (or “bucket”) for each unique and/or similar value of the fact with the same attribute in the merge candidates. Thus, if the grouping is performed based on the name fact, and there are two merge candidates, one containing the name “George Bush” and the other containing the names “Bill Clinton” and “William Jefferson Clinton,” the grouping module 310 creates three groups. An object having multiple name facts is placed in multiple groups. For example, the object containing the names “Bill Clinton” and “William Jefferson Clinton” is placed in the “Bill Clinton” group and in the “William Jefferson Clinton” group. One embodiment of the grouping module 310 accounts for synonyms (e.g., “Bill” and “William”) and other minor variations (such as occurrences of middle names, “jr.,” “Inc.” and other such terms in strings) when forming the groups.
Furthermore, other embodiments of the grouping module 310 group based on facts or aspects other than names. For example, the grouping module 310 can group based on facts such as “date of birth,” “height,” “weight,” and both height and weight. Likewise, the grouping module 310 can group based on the source domain from which the object was extracted. Again, an embodiment of the grouping module 310 accounts for synonyms, equivalencies, and minor variations when forming the groups. For example, objects containing the facts “height: 91 inches” and “height: 2.3 meters” can be grouped together because the heights are very similar even though the facts are not identical. One embodiment groups by height rounded to the nearest five centimeters to accommodate small errors in measurement.
An object comparison module 312 compares the merge candidates within each group. In one embodiment, the object comparison module 312 operates on two objects at a time and compares every pair of objects in the group. The object comparison module 312 computes a similarity value that measures the similarity between the two objects. There are a variety of ways to compute the similarity value. The similarity value computation utilized by one embodiment of the object comparison module 312 is described herein. Other embodiments can use different and/or additional computations.
When comparing a pair of objects within a group, an embodiment of the object comparison module 312 determines whether the objects share any or all of the following features: 1) the entities share at least one type; 2) a term frequency/inverse document frequency (TF/IDF) metric exceeds a specified threshold; 3) the two objects came from the same web page or other electronic document; 4) the web link graph for the objects is greater than zero and less than a specified threshold; or 5) the objects share an uncommon fact. For each feature, the object comparison module 312 assigns a similarity weight between zero and one inclusive indicating whether the objects share the feature, and the extent of the sharing.
Turning now to the individual features, each object has zero or more special “type” facts that describe the type of entity represented by the object. Examples of types include “person,” “first baseman,” “car,” “mountain,” and “mathematical concept.” An object can have multiple types depending upon the facts present in the document on which the object is based. For example, an object with the name “George Bush” can have a type “person” and a type “ship” if the underlying document describes both President Bush and the aircraft carrier named after him. The object comparison module 312 determines whether the two objects being compared have at least one type fact in common. In one embodiment, the object comparison module 312 computes a similarity weight of zero or one in response to the comparison, with a weight of one indicating that the objects share a common type.
The TF/IDF metric measures the similarity of text contained in the two objects being compared relative to the similarity with text in the overall corpus (e.g., the web at large or the entire set of objects in the repository 115). In other words, the TF/IDF metric measures whether the two objects have more words in common than one would expect given the frequency that the words are used in the relevant domain. The TF/IDF metric is useful because many objects have facts with values that are text strings. For example, two objects about the first President George Bush might each have a fact that contains a lengthy text string describing his biography. The TF/IDF metric comparison would determine that both facts use words like “baseball,” “Yale,” “Barbara,” “President,” etc. more often than these words are used by objects, facts, or documents describing other entities. In one embodiment, the object comparison module 312 computes a similarity weight between zero and one inclusive for the TF/IDF metric, where a weight of zero indicates that the facts of the objects have a very low TF/IDF similarity and a weight of one indicates that the facts of the objects have a very high TF/IDF similarity.
If two objects are derived from the same web page or other electronic document, then there is a strong possibility that the objects are associated with the same entity. Accordingly, an embodiment of the object comparison module 312 computes a similarity weight of one if the objects are from the same document, and a weight of zero if the objects are derived from different documents. Similarly, if the objects are not derived from the same web page, an embodiment of the object comparison module 312 determines the web link graph distance between the pages from which the objects are derived. The web link graph distance measures the number of hypertext links one must follow to reach one page from the other. For example, if a first page contains a hypertext link directly to a second page, the web link graph distance between the pages is one. An embodiment of the object comparison module 312 computes a similarity weight of zero if the web link graph distance exceeds a specified threshold, and a weight of one if the web link graph distance is below the threshold. The web link graph distance threshold in one embodiment is two.
If two objects share an uncommon fact, then the objects might be associated with the same entity. For example, if two objects share the fact (height, 8850 Meters), they might both be associated with Mount Everest. Likewise, if two objects of type “person” share the fact (height, 7′ 5″), they are likely associated with the same person because this is an unusually tall height. An embodiment of the object comparison module 312 computes a similarity weight of zero if the pair of objects do not share an uncommon fact, and a weight of one if the objects share such a fact.
In one embodiment, the object comparison module 312 uses statistical techniques in order to identify facts that are “uncommon” for purposes of object comparison. The object comparison module 312 computes a normal distribution of values for a fact, and considers values that fall outside of a given range, e.g., two standard deviations from the mean, as “uncommon.” For example, the object comparison module 312 can create a histogram of the values of “height” facts from objects of type “person.” This histogram will probably show a normal distribution (bell curve) centered at approximately 170 cm (assuming that both genders are included). Heights on the shoulders of the bell curve are “uncommon.” In one embodiment, the object comparison module 312 and/or another module analyzes facts in the repository 115 in order to identify common and uncommon values. The facts can be analyzed as part of a preprocessing step, and/or analyzed in real time when two objects are being compared. An administrator or other entity can designated the facts and/or types of facts to be analyzed.
When comparing a pair of objects within a group, an embodiment of the object comparison module 312 also determines whether there is evidence that the objects are not associated with the same entity. In one embodiment, such evidence exists if the objects: 1) have differing singleton attributes; 2) were added by trusted agents; or 3) have type conflicts. These types of evidence are described in more detail below. In one embodiment, the object comparison module 312 applies a negative infinity similarity weight to a pair of objects that contain any evidence that they are not associated with the same entity. This similarity weight ensures that the two objects are not merged.
A singleton attribute can take on only one value for a given entity. For example, the “date of birth” attribute for a person should only have a single value. Similarly, the “number of pages” for a book, “height” of a mountain, “length” of a song, and “chemical formula” for a drug should each have only a single value. If the pair of objects being compared have different singleton attributes, an embodiment of the object comparison module 312 assumes that the objects must be associated with different entities and thus assigns a negative infinity similarity weight to the pair. The singleton attributes can be identified by human users or machine-learned through an analysis of the objects in the repository 115.
Some objects are added by trusted agents. A trusted agent is a tool that is specifically designed to create objects based on facts contained in a specific set of electronic documents known to be mutually unique. For example, an agent can be designed to create objects and/or facts based on the contents of web pages from sites like Wikipedia, the CIA World Factbook, or another web site that enforces this uniqueness. These sites generally do not contain multiple web pages for the same entity. If the two objects under comparison were derived from the same web site and added by the same trusted agent, an embodiment of the object comparison module 312 assumes that the agent properly distinguished among the entities associated with the objects. Accordingly, the object comparison module 312 assigns a negative infinity similarity weight to the pair to ensure that they are not merged even if the objects contain similar facts.
A type conflict exists if a pair of objects have facts indicating that they are of mutually exclusive types. For example, if one object has a type fact indicating that it is associated with a person, and another object has a type fact indicating that it is associated with a ship, then the objects are likely not associated with the same entity. The object comparison module thus assigns a negative infinity similarity weight to the pair to prevent them from being merged.
The object comparison module 312 combines the similarity weights to produce the similarity value and stores this final value in association with the pair of objects. In one embodiment, the object comparison module 312 assigns a similarity value of −0.1 or another small negative value if the object pair shares no features in order to prevent a merger of the objects. In addition, the object comparison module 312 assigns a similarity value of zero to pairs of objects that are in different groups and therefore cannot share features. Thus, the similarity value assigned to a pair of objects in this embodiment is either negative infinity or within the range between −0.1 (no shared features) and five (all features shared to maximum possible extent) inclusive. A positive similarity value indicates that there is evidence for a merge, a negative value indicates that there is evidence against a merge, and a zero value is non-determinative.
A graph generation module 314 creates graphs describing the comparisons performed by the object comparison module 312. In one embodiment, the graph generation module 314 creates graphs linking sets of possibly-related objects across all of the groups. Staying with the name-based group embodiment, recall that an object can have multiple names and therefore can be placed in multiple groups. Since an object is paired with every other object in its group, and the object can also be a member of multiple groups, the relatedness graph for the object can span multiple name-based groups. In one embodiment, the graph generation module 314 creates graphs having nodes representing objects and edges between the nodes representing possible relationships between the objects. The weight of an edge between two nodes is the similarity value for the pair of objects represented by the nodes.
While this description refers to the output of the graph generation module 314 as a “graph,” embodiments of the module can generate outputs that are not technically graphs. There are many ways to represent graphs and related data structures within a computer, and embodiments of the graph generation module 314 can use any such technique. For example, an embodiment of the graph generation module 314 can use a table to represent the relationships of the objects. As used herein, the term “graph” is intended to cover other, non-graph-based, representations of the objects.
A merging module 316 analyzes the graphs produced by the graph generation module 314 and merges objects that are likely associated with the same entity. In one embodiment, the merging module 316 establishes a threshold for the similarity value. Pairs of objects having a similarity value above the threshold are considered associated with the same entity and are merged. In one embodiment, the threshold is zero.
In one embodiment, the merging module 316 performs a hierarchical agglomerative clustering on each graph in order to merge nodes and the objects represented by the nodes. For a given graph, the merging module 316 identifies the edge having the greatest similarity value that exceeds the threshold and merges the nodes connected by the edge (and the associated objects). Then, then merging module 316 again finds the greatest similarity value and merges the associated nodes/objects. This merging process continues until either all nodes in the graph are merged or the similarity values between all of the remaining nodes are below the threshold.
In one embodiment, the similarity values between a merged object and other objects in the graph are recomputed upon each merge. Assume there are two objects O1 and O2 that are to be merged, and a third object O3 that is in the same graph. When O1 and O2 are merged, the similarity values between O1 and O3 and between O2 and O3 are summed to produce the similarity value between the O1/O2 and O3 pair. This summing causes an accelerating effect where O3 is likely to be either a lot more or a lot less similar to O1/O2 than it was to either O1 or O2 individually.
An embodiment of the merging module 316 merges two or more objects by combining the facts of the objects into a single object. In one embodiment, identical facts contained in both objects are merged into a single fact. Different facts in the objects are retained in the merged object. The merging module 316 stores the merged object in the repository 115 in place of, or in addition to, the objects from which it was formed.
Initially, the object merge engine 300 groups 410 a set of merge candidate objects. The engine 300 compares each pair of objects in each group. For a given pair of objects, the engine 300 compares 412 the objects for evidence in favor of merging them. For example, both objects containing the same uncommon fact can be evidence in favor of merging. In addition, the engine 300 compares 414 the objects for evidence against merging them. For example, both objects being added by the same trusted agent can be evidence against merging them. The object merge engine 300 computes 416 a similarity value for each pair of objects based on the comparisons. Further, the engine 300 graphs 418 the objects based on their group relationships and similarity values. The engine 300 performs hierarchical agglomerative clustering on the graphs and merges 420 objects that have a similarity value exceeding a threshold. The merged objects are stored in the repository 115.
In one embodiment, janitors 110 associated with the merge engine 300 and/or other entities benefit from the first merging round. These janitors 110 clean 422 the post-merge data in the repository 115 to the point where additional merging is possible in a subsequent round. These optional subsequent rounds are shown in
To understand the operation of the object merge engine 300 described above, consider an example using the following four merge candidate objects:
Object 1
name: Bill Clinton
type: person
date of birth: 19 Aug. 1946
wife: Hillary Clinton
Object 2
name: William Jefferson Clinton
type: person
place of birth: Hope, Ark.
Object 3
name: Bill Clinton
name: William Jefferson Clinton
type: person
date of birth: Aug. 19, 1946
Object 4
name: Bill Clinton
type: person
date of birth: Mar. 12, 1793.
These objects collectively have two different values for the name fact: “Bill Clinton” and “William Jefferson Clinton.” Thus, the object merge engine 300 creates a group for each name and assigns the objects to the groups as follows:
Group “Bill Clinton:” Objects 1, 3, and 4
Group “William Jefferson Clinton”: Objects 2 and 3.
The object merge engine 300 compares each pair of objects in a group for evidence in favor of a merge and evidence against a merge, and computes a similarity value based on the evidence. Assume for this example that a pair with only evidence in favor of a merge receives a similarity value of “1” while a pair with evidence against a merge receives a similarity value of “−∞.” The similarity values for the pairs in the groups are:
(1, 3): 1
(1, 4): −∞
(3, 4): −∞
(2, 3): 1
Object pairs (1,3) and (2,3) each receive a similarity value of “1” because they share a common type. Pairs (1,4) and (3,4) each receive a similarity value of “−∞” because the attribute “date of birth” is singleton and object four has a date of birth fact with the value of “Mar. 12, 1793.” This different singleton attribute constitutes evidence against a merger with objects one or three.
The object merge engine 300 performs hierarchical agglomerative clustering on the graph of
In other embodiments, the object merge engine 300 uses different and/or additional techniques to merge objects. These techniques can be used on their own, or in combination with the techniques described above.
In one such technique, the object comparison module 312 compares each pair of objects in a group to determine whether one object is a subset of the other. A first object is a subset of a second object if all of the facts in the first object are also found in the second object. Once the subset relationships are determined, the graph generation module 314 builds directed graphs describing the subset relationships of all of the objects from all of the groups. A first node representing a first object having an edge directed to a second node representing a second object indicates that the first object is a subset of the second object.
In the directed graph, a node representing an object that is a superset of other objects and not a subset itself will constitute a graph sink. That is, the node will have one or more edges entering the node but no edges leaving the node. If the node representing an object in the graph is connected by directed edges to exactly one sink node, the merging module 316 merges the object into the object represented by the sink node. Further, if the node representing the object in the graph is connected to more than one sink node, then the merging module 316 deletes the object from the repository 115 because it adds no information to the repository and is ambiguous.
In some embodiments, the techniques described herein are performed on entities other than objects. For example, the techniques can be applied directly to web pages or other electronic documents. In such an embodiment, the output is a cluster of web pages of facts.
Reference in the specification to “one embodiment” or to “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiments is included in at least one embodiment of the invention. The appearances of the phrase “in one embodiment” in various places in the specification are not necessarily all referring to the same embodiment.
Some portions of the above are presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is here, and generally, conceived to be a self-consistent sequence of steps (instructions) leading to a desired result. The steps are those requiring physical manipulations of physical quantities. Usually, though not necessarily, these quantities take the form of electrical, magnetic or optical signals capable of being stored, transferred, combined, compared and otherwise manipulated. It is convenient at times, principally for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like. Furthermore, it is also convenient at times, to refer to certain arrangements of steps requiring physical manipulations of physical quantities as modules or code devices, without loss of generality.
It should be borne in mind, however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise as apparent from the following discussion, it is appreciated that throughout the description, discussions utilizing terms such as “processing” or “computing” or “calculating” or “determining” or “displaying” or “determining” or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system memories or registers or other such information storage, transmission or display devices.
Certain aspects of the present invention include process steps and instructions described herein in the form of an algorithm. It should be noted that the process steps and instructions of the present invention can be embodied in software, firmware or hardware, and when embodied in software, can be downloaded to reside on and be operated from different platforms used by a variety of operating systems.
The present invention also relates to an apparatus for performing the operations herein. This apparatus may be specially constructed for the required purposes, or it may comprise a general-purpose computer selectively activated or reconfigured by a computer program stored in the computer. Such a computer program may be stored in a computer readable storage medium, such as, but is not limited to, any type of disk including floppy disks, optical disks, CD-ROMs, magnetic-optical disks, read-only memories (ROMs), random access memories (RAMs), EPROMs, EEPROMs, magnetic or optical cards, application specific integrated circuits (ASICs), or any type of media suitable for storing electronic instructions, and each coupled to a computer system bus. Furthermore, the computers referred to in the specification may include a single processor or may be architectures employing multiple processor designs for increased computing capability.
The algorithms and displays presented herein are not inherently related to any particular computer or other apparatus. Various general-purpose systems may also be used with programs in accordance with the teachings herein, or it may prove convenient to construct more specialized apparatus to perform the required method steps. The required structure for a variety of these systems will appear from the description below. In addition, the present invention is not described with reference to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the present invention as described herein, and any references below to specific languages are provided for disclosure of enablement and best mode of the present invention.
While the invention has been particularly shown and described with reference to a preferred embodiment and several alternate embodiments, it will be understood by persons skilled in the relevant art that various changes in form and details can be made therein without departing from the spirit and scope of the invention.
Finally, it should be noted that the language used in the specification has been principally selected for readability and instructional purposes, and may not have been selected to delineate or circumscribe the inventive subject matter. Accordingly, the disclosure of the present invention is intended to be illustrative, but not limiting, of the scope of the invention, which is set forth in the following claims.
Number | Name | Date | Kind |
---|---|---|---|
5010478 | Deran | Apr 1991 | A |
5133075 | Risch | Jul 1992 | A |
5347653 | Flynn et al. | Sep 1994 | A |
5440730 | Elmasri et al. | Aug 1995 | A |
5475819 | Miller et al. | Dec 1995 | A |
5519608 | Kupiec | May 1996 | A |
5560005 | Hoover et al. | Sep 1996 | A |
5574898 | Leblang et al. | Nov 1996 | A |
5680622 | Even | Oct 1997 | A |
5694590 | Thuraisingham et al. | Dec 1997 | A |
5701470 | Joy et al. | Dec 1997 | A |
5717911 | Madrid et al. | Feb 1998 | A |
5717951 | Yabumoto | Feb 1998 | A |
5778378 | Rubin | Jul 1998 | A |
5787413 | Kauffman et al. | Jul 1998 | A |
5793966 | Amstein et al. | Aug 1998 | A |
5802299 | Logan et al. | Sep 1998 | A |
5815415 | Bentley et al. | Sep 1998 | A |
5819210 | Maxwell, III et al. | Oct 1998 | A |
5819265 | Ravin et al. | Oct 1998 | A |
5822743 | Gupta et al. | Oct 1998 | A |
5826258 | Gupta et al. | Oct 1998 | A |
5909689 | Van Ryzin | Jun 1999 | A |
5920859 | Li | Jul 1999 | A |
5943670 | Prager | Aug 1999 | A |
5956718 | Prasad et al. | Sep 1999 | A |
5974254 | Hsu | Oct 1999 | A |
5987460 | Niwa et al. | Nov 1999 | A |
6018741 | Howland et al. | Jan 2000 | A |
6044366 | Graffe et al. | Mar 2000 | A |
6052693 | Smith et al. | Apr 2000 | A |
6078918 | Allen et al. | Jun 2000 | A |
6112203 | Bharat et al. | Aug 2000 | A |
6112210 | Nori et al. | Aug 2000 | A |
6122647 | Horowitz et al. | Sep 2000 | A |
6134555 | Chadha et al. | Oct 2000 | A |
6138270 | Hsu | Oct 2000 | A |
6212526 | Chaudhuri et al. | Apr 2001 | B1 |
6240546 | Lee et al. | May 2001 | B1 |
6263328 | Coden et al. | Jul 2001 | B1 |
6285999 | Page | Sep 2001 | B1 |
6289338 | Stoffel et al. | Sep 2001 | B1 |
6311194 | Sheth et al. | Oct 2001 | B1 |
6327574 | Kramer et al. | Dec 2001 | B1 |
6349275 | Schumacher et al. | Feb 2002 | B1 |
6377943 | Jakobsson | Apr 2002 | B1 |
6397228 | Lamburt et al. | May 2002 | B1 |
6473898 | Waugh et al. | Oct 2002 | B1 |
6502102 | Haswell et al. | Dec 2002 | B1 |
6556991 | Borkovsky | Apr 2003 | B1 |
6567936 | Yang et al. | May 2003 | B1 |
6572661 | Stern | Jun 2003 | B1 |
6584464 | Warthen | Jun 2003 | B1 |
6606625 | Muslea et al. | Aug 2003 | B1 |
6606659 | Hegli et al. | Aug 2003 | B1 |
6609123 | Cazemier et al. | Aug 2003 | B1 |
6656991 | Staccione et al. | Dec 2003 | B2 |
6665659 | Logan | Dec 2003 | B1 |
6665666 | Brown et al. | Dec 2003 | B1 |
6665837 | Dean et al. | Dec 2003 | B1 |
6745189 | Schreiber | Jun 2004 | B2 |
6754873 | Law et al. | Jun 2004 | B1 |
6799176 | Page | Sep 2004 | B1 |
6804667 | Martin | Oct 2004 | B1 |
6820081 | Kawai et al. | Nov 2004 | B1 |
6823495 | Vedula et al. | Nov 2004 | B1 |
6850896 | Kelman et al. | Feb 2005 | B1 |
6886005 | Davis | Apr 2005 | B2 |
6901403 | Bata et al. | May 2005 | B1 |
6957213 | Yuret | Oct 2005 | B1 |
6963880 | Pingte et al. | Nov 2005 | B1 |
7003522 | Reynar et al. | Feb 2006 | B1 |
7003719 | Rosenoff et al. | Feb 2006 | B1 |
7020662 | Boreham et al. | Mar 2006 | B2 |
7051023 | Kapur et al. | May 2006 | B2 |
7080073 | Jiang et al. | Jul 2006 | B1 |
7080085 | Choy et al. | Jul 2006 | B1 |
7143099 | Lecheler-Moore et al. | Nov 2006 | B2 |
7146536 | Bingham, Jr. et al. | Dec 2006 | B2 |
7162499 | Lees et al. | Jan 2007 | B2 |
7194380 | Barrow et al. | Mar 2007 | B2 |
7197449 | Hu et al. | Mar 2007 | B2 |
7277879 | Varadarajan | Oct 2007 | B2 |
7305380 | Hoelzle et al. | Dec 2007 | B1 |
7363312 | Goldsack | Apr 2008 | B2 |
7472182 | Young et al. | Dec 2008 | B1 |
7483829 | Murakami et al. | Jan 2009 | B2 |
7493317 | Geva | Feb 2009 | B2 |
7797282 | Kirshenbaum et al. | Sep 2010 | B1 |
20020038307 | Obradovic et al. | Mar 2002 | A1 |
20020042707 | Zhao et al. | Apr 2002 | A1 |
20020065845 | Naito et al. | May 2002 | A1 |
20020073115 | Davis | Jun 2002 | A1 |
20020083039 | Ferrari et al. | Jun 2002 | A1 |
20020087567 | Spiegler et al. | Jul 2002 | A1 |
20020147738 | Reader | Oct 2002 | A1 |
20020169770 | Kim et al. | Nov 2002 | A1 |
20020178448 | Te Kiefte et al. | Nov 2002 | A1 |
20020194172 | Schreiber | Dec 2002 | A1 |
20030018652 | Heckerman et al. | Jan 2003 | A1 |
20030058706 | Okamoto et al. | Mar 2003 | A1 |
20030078902 | Leong et al. | Apr 2003 | A1 |
20030120644 | Shirota | Jun 2003 | A1 |
20030120675 | Stauber et al. | Jun 2003 | A1 |
20030126102 | Borthwick | Jul 2003 | A1 |
20030149567 | Schmitz et al. | Aug 2003 | A1 |
20030154071 | Shreve | Aug 2003 | A1 |
20030177110 | Okamoto et al. | Sep 2003 | A1 |
20030182310 | Charnock et al. | Sep 2003 | A1 |
20030195877 | Ford et al. | Oct 2003 | A1 |
20030196052 | Bolik et al. | Oct 2003 | A1 |
20040003067 | Ferrin | Jan 2004 | A1 |
20040024739 | Copperman et al. | Feb 2004 | A1 |
20040064447 | Simske et al. | Apr 2004 | A1 |
20040088292 | Dettinger et al. | May 2004 | A1 |
20040107125 | Guheen et al. | Jun 2004 | A1 |
20040122844 | Malloy et al. | Jun 2004 | A1 |
20040123240 | Gerstl et al. | Jun 2004 | A1 |
20040128624 | Arellano et al. | Jul 2004 | A1 |
20040143600 | Musgrove et al. | Jul 2004 | A1 |
20040153456 | Charnock et al. | Aug 2004 | A1 |
20040167870 | Wakefield et al. | Aug 2004 | A1 |
20040177015 | Galai et al. | Sep 2004 | A1 |
20040199923 | Russek | Oct 2004 | A1 |
20040240542 | Yeredor et al. | Dec 2004 | A1 |
20040255237 | Tong | Dec 2004 | A1 |
20040260979 | Kumai | Dec 2004 | A1 |
20040267700 | Dumais et al. | Dec 2004 | A1 |
20050076012 | Manber et al. | Apr 2005 | A1 |
20050086211 | Mayer | Apr 2005 | A1 |
20050086222 | Wang et al. | Apr 2005 | A1 |
20050097150 | McKeon et al. | May 2005 | A1 |
20050125311 | Chidiac et al. | Jun 2005 | A1 |
20050149576 | Marmaros et al. | Jul 2005 | A1 |
20050149851 | Mittal | Jul 2005 | A1 |
20050187923 | Cipollone | Aug 2005 | A1 |
20050240615 | Barsness et al. | Oct 2005 | A1 |
20050256825 | Dettinger et al. | Nov 2005 | A1 |
20060026122 | Hurwood et al. | Feb 2006 | A1 |
20060036504 | Allocca et al. | Feb 2006 | A1 |
20060041597 | Conrad et al. | Feb 2006 | A1 |
20060047838 | Chauhan | Mar 2006 | A1 |
20060053171 | Eldridge et al. | Mar 2006 | A1 |
20060053175 | Gardner et al. | Mar 2006 | A1 |
20060074824 | Li | Apr 2006 | A1 |
20060074910 | Yun et al. | Apr 2006 | A1 |
20060085465 | Nori et al. | Apr 2006 | A1 |
20060136585 | Mayfield et al. | Jun 2006 | A1 |
20060143227 | Helm et al. | Jun 2006 | A1 |
20060143603 | Kalthoff et al. | Jun 2006 | A1 |
20060152755 | Curtis et al. | Jul 2006 | A1 |
20060238919 | Bradley | Oct 2006 | A1 |
20060248045 | Toledano et al. | Nov 2006 | A1 |
20060248456 | Bender et al. | Nov 2006 | A1 |
20060253418 | Charnock et al. | Nov 2006 | A1 |
20060288268 | Srinivasan et al. | Dec 2006 | A1 |
20060293879 | Zhao et al. | Dec 2006 | A1 |
20070005593 | Self et al. | Jan 2007 | A1 |
20070005639 | Gaussier et al. | Jan 2007 | A1 |
20070016890 | Brunner et al. | Jan 2007 | A1 |
20070038610 | Omoigui | Feb 2007 | A1 |
20070073768 | Goradia | Mar 2007 | A1 |
20070094246 | Dill et al. | Apr 2007 | A1 |
20070130123 | Majumder | Jun 2007 | A1 |
20070143317 | Hogue et al. | Jun 2007 | A1 |
20070150800 | Betz et al. | Jun 2007 | A1 |
20070198480 | Hogue et al. | Aug 2007 | A1 |
20070198481 | Hogue et al. | Aug 2007 | A1 |
20070203867 | Hogue et al. | Aug 2007 | A1 |
20070271268 | Fontoura et al. | Nov 2007 | A1 |
20080071739 | Kumar et al. | Mar 2008 | A1 |
20080127211 | Belsey et al. | May 2008 | A1 |
20090006359 | Liao | Jan 2009 | A1 |
Number | Date | Country |
---|---|---|
5-174020 | Jul 1993 | JP |
WO 0127713 | Apr 2001 | WO |
WO 2004114163 | Dec 2004 | WO |
WO 2006104951 | Oct 2006 | WO |