WO2012115965A1 - Empreintes digitales d'entités - Google Patents

Empreintes digitales d'entités Download PDF

Info

Publication number
WO2012115965A1
WO2012115965A1 PCT/US2012/025942 US2012025942W WO2012115965A1 WO 2012115965 A1 WO2012115965 A1 WO 2012115965A1 US 2012025942 W US2012025942 W US 2012025942W WO 2012115965 A1 WO2012115965 A1 WO 2012115965A1
Authority
WO
WIPO (PCT)
Prior art keywords
entity
attribute
entities
fingerprint
identifier
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/US2012/025942
Other languages
English (en)
Inventor
Hassan H. MALIK
Vikas S. BHARDWAJ
Ian MACGILLIVRAY
Christy HATCH
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Thomson Reuters Global Resources ULC
Original Assignee
Thomson Reuters Global Resources ULC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US13/107,665 external-priority patent/US9495635B2/en
Priority claimed from US13/213,324 external-priority patent/US9292545B2/en
Application filed by Thomson Reuters Global Resources ULC filed Critical Thomson Reuters Global Resources ULC
Priority to CN201280019646.3A priority Critical patent/CN103827895B/zh
Priority to EP12707690.9A priority patent/EP2678809A1/fr
Publication of WO2012115965A1 publication Critical patent/WO2012115965A1/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/3332—Query translation
    • G06F16/3334—Selection or weighting of terms from queries, including natural language queries

Definitions

  • This disclosure relates to entity fingerprints, and more particularly to systems and techniques for computing and analyzing entity fingerprints.
  • Entity-centric models are traditionally built with strong reference to structured content: a database of people's personal details or of geographical information, for example,
  • the systems and techniques provide an entity-based information analysis and content aggregation platform that uses heterogeneous data sources to construct and maintain an ecosystem around tangible and logical entities such as, but not limited to, organizations, people, products, industries, geographies, commodities, financial indicators, economic indicators, events, topics, subject codes, unique identifiers, social tags, industry terms, general terms, metadata elements, classification codes, and combinations thereof.
  • Entities are represented as vertices in a directed graph, and edges are generated using entity co-occurrences in unstmctured documents and supervised information from structured data sources. Significance scores for the edges are computed using a technique that combines supervised, unsupervised and temporal factors into a single score. Important entity attributes from the structured content and the entity neighborhood in the graph are automatically summarized as the entity fingerprint. Entities may be compared to one another based on similarity of their entity fingerprints.
  • An interactive user interface is also disclosed that provides exploratory access to the graph and supports decision support processes.
  • Various aspects of the invention relate to computing and analyzing entity fingerprints.
  • a method of analyzing an entity includes identifying a first entity from a plurality of entities, each of the plurality of entities associated with an entity fingerprint comprising at least one assigned attribute, comparing the entity fingerprint associated with the first entity to each of the entity fingerprints associated with the plurality of entities, and identifying at least one similar entity to the first entity from the plurality of entities based on the fingerprint comparisons.
  • the method also includes generating a signal associated with the at least one identified similar entity and the first entity, and transmitting the signal.
  • identifying the first entity comprises receiving a unique identifier associated with the first entity.
  • the at least one assigned attribute associated with each entity fingerprint is based on a relevancy of the assigned attribute to the associated entity over a particular time interval.
  • the assigned attribute may also be associated with a weight value indicating a relevancy of the assigned attribute to the associated entity as compared to other assigned attributes of the entity.
  • the method further includes ranking a plurality of similar entities in order of similarity to the first entity, generating a second signal associated with the ranking; and transmitting the second signal in response to a request.
  • the method may also include providing a graphical user interface for viewing at least one attribute of the entity fingerprint, at least one attribute of at least one similar entity, at least one document related to the first entity, at least one document related to the first entity and the at least one similar entity, or combination thereof.
  • the method further includes comparing entity fingerprints by computing a first weighted value for each unique attribute-type associated with the first entity by applying an aggregate function to attribute weights associated with attributes corresponding to each unique attribute-type of the first entity, computing a second weighted value for each unique attribute-type associated with the at least one similar entity by applying the aggregate function to attribute weights associated with attributes corresponding to each unique attribute type of the similar entity, and normalizing the attributes associated with each attribute-type of the first entity using the computed first weighted value and of the second entity using the computed second weighted value, A similarity score is then computed for the first and second fingerprints by applying a similarity function to the normalized attributes of the first entity and the similar entity.
  • the aggregate function may be, but is not limited to an average function, a minimum function, a maximum function, and the like
  • the similarity function may be, but is not limited to a cosine similarity, a Euclidean distance, a Manhattan distance, and the like.
  • a method of computing an entity fingerprint includes reading data representative of a set of attributes associated with an entity for which an entity fingerprint is desired, and processing the data in accordance with at least one fingerprint extraction algorithm.
  • the fingerprint extraction algorithm computing the entity fingerprint by representing a plurality of entities including the entity as vertices of a directed graph, generating at least one edge of the directed graph using entity co-occurrences in a set of unstructured documents and structured information sources, and summarizing the set of attributes associated with the entity for which the entity fingerprint is desired using the structured information sources and an entity
  • processing the data further includes initializing a data structure for association with the entity, the data structure including at least one attribute identifier, associated attribute-type identifier, associated attribute value, and associated attribute weight, comparing the entity with a set of pre-defined entities, each of the pre-defined entities optionally associated with an attribute identifier, attribute-type identifier, and attribute value, and importing values for the data structure based on the comparison.
  • the method also includes identifying information sources for association with the data structure from the set of structured information sources, Each of the set of structured information sources is optionally associated with an attribute identifier, an attribute-type identifier, an attribute value, and an information-source identifier, importing values for the data structure based on the identification of the information sources comprising information associated with the entity, identifying vertices of the directed graph that connect to the entity and comprise the entity neighborhood. Values are imported into the data structure based on at least one of the vertices and set of edges of the directed graph that connect to the entity and comprise the entity neighborhood.
  • the method further includes identifying documents for association with the entity based on metadata associated with each document of a set of documents, importing values for the data structure based on the identified documents, and associating the data structure with the entity.
  • the method further includes storing a reference to each of the identified documents in at least one of the vertices and the set of edges of the directed graph.
  • the at least one associated attribute is determined from the directed graph and the at least one associated attribute weight is determined from a set of pre-defined weights.
  • Each of the set of pre-defined weights includes a weight value and at least one of an information-source identifier, an attribute-type identifier, an obtained-from-neighborhood identifier and a validated identifier.
  • each edge of the directed graph optionally includes an attribute identifier, an attribute-type identifier, and an attribute value.
  • the method includes comparing a significant score attribute associated with an edge of the directed graph to a user-defined threshold value, and importing values for the data structure based on the comparison.
  • the method may also include comparing the metadata associated with each document of the set of documents to an entity identifier associated with the entity, and importing values for the data structure based on the comparison.
  • the method may include comparing metadata associated with each document of the set of documents to user-defined criteria, the user-defined criteria identifying at least one peer entity to the entity and not including the entity, and importing values for the data structure based on the comparison.
  • the method may also include, for example, ranking one of entities and metadata optionally associated with each document of the set of documents for each attribute type associated therewith, and selecting a subset of documents for association with the data structure based on the ranking.
  • FIG. 1 is a schematic of an exemplary computer-based system for computing and comparing an entity fingerprint.
  • FIG. 2 illustrates an exemplary method for generating an entity fingerprint.
  • FIG. 3 illustrates an exemplary method for comparing entity fingerprints.
  • FIG. 4 is a schematic of an exemplary directed graph.
  • FIG. 5 illustrates an example graphical user interface for displaying an entity fingerprint.
  • FIG. 6 illustrates an example query frame of the graphical user interface shown in FIG. 5.
  • FIGS. 7-8 show example matrix frames of the graphical user interface shown in FIG. 5.
  • FIG. 9 illustrates a news overlay of the graphical user interface shown in FIG. 5.
  • FIG. 10 illustrates a headline frame of the graphical user interface shown in FIG. 5.
  • FIG. 11 illustrates an example entity fingerprint display.
  • FIG. 12 illustrates a schematic side-by-side display.
  • FIG. 13 illustrates an example side-by-side display.
  • the tangible and logical entities may include, but are not limited to, organizations, people, products, industries, geographies, commodities, financial indicators, economic indicators, events, topics. subject codes, unique identifiers, social tags, industry terms, general terms, metadata elements, classification codes, technologies, facilities, media, and combinations thereof.
  • entities are represented as vertices in a directed graph, and edges of the directed graph are generated using entity co-occurrences in unstructured documents and structured information available from structured data sources. Vertices and edges of the directed graph may include properties and references to relevant documents, and a single edge of the graph may represent multiple types of relationships between entities.
  • some entities are validated by mapping each entity to a known entity from a structured data source (which may be human-maintained) while other entities are discovered and are solely obtained from unstructured data sources.
  • Significance scores for the edges of the graph are computed using a technique that combines supervised, unsupervised and temporal factors into a single score.
  • the technique is disclosed in U.S. Patent Application No.: 13/107,665 filed May 13, 2011, entitled 'Association Significance', the content of which is incorporated herein in its entirety.
  • Entity attributes derived from the structured content and the entity neighborhood in the graph are automatically summarized as an entity fingerprint. While the present invention is discussed in connection with a directed graph, it will be appreciated that the present invention is not limited to a directed graph implementation, and that other computer-implemented data structures capable of modeling entity relationships may be used with the present invention, such as a mixed graph and a multigraph. Further, in one embodiment, the system also includes an interactive user interface that provides exploratory access to the graph and supports common business use cases, such as determining entity similarity and comparing entities.
  • FIG. 1 an example of a suitable computing system 10 within which embodiments of the present invention may be implemented is disclosed. The computing system 10 is only one example and is not intended to suggest any limitation as to the scope of use or functionality of the invention. Neither should the computing system 10 be interpreted as having any dependency or requirement relating to any one or combination of illustrated components.
  • the present invention is operational with numerous other general purpose or special purpose computing consumer electronics, network PCs, minicomputers, mainframe computers, laptop computers, as well as distributed computing environments that include any of the above systems or devices, and the like.
  • the invention may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer.
  • program modules include routines, programs, objects, components, data structures, loop code segments and constructs, etc. that perform particular tasks or implement particular abstract data types.
  • the invention can be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network.
  • program modules are located in both local and remote computer storage media including memory storage devices. Tasks performed by the programs and modules are described below and with the aid of figures.
  • processor executable instructions which can be written on any form of a computer readable media.
  • the programming language used is Java.
  • the directed graph is loaded into a computer memory as a runtime Java object, and is persisted to a data store, such as a file system, as a serialized Java object.
  • Information needed to generate the directed graph is stored in one or more structured data sources, such as a relational database.
  • the following software libraries may be used to implement components of the system. These include, but are not limited to, Apache Lucene 3.0.1, text search engine library, and Apache XML Beans 2,4, Further, to generate Java objects from XML schemas, the following may also be utilized: Jetty 6.01, a HTTP Web server and client; and a fast XML parser, such as VTD-XML.
  • the system 10 includes a server device 12 configured to include a processor 14, such as a central processing unit ('CPU'), random access memory ('RAM') 16, one or more input-output devices 18, such as a display device (not shown) and keyboard (not shown), and non- volatile memory 20, all of which are interconnected via a common bus 22 and controlled by the processor 14.
  • a processor 14 such as a central processing unit ('CPU'), random access memory ('RAM') 16
  • input-output devices 18 such as a display device (not shown) and keyboard (not shown)
  • non- volatile memory 20 all of which are interconnected via a common bus 22 and controlled by the processor 14.
  • the non- volatile memory 20 is configured to include a fingerprint extraction module 26 for computing and comparing entity fingerprints to one another.
  • a fingerprint extraction module 26 for computing and comparing entity fingerprints to one another.
  • 'fingerprint' refers to an abstract representation of an entity based on a number of its attributes and/or characteristics. Once a fingerprint is computed for an entity, the entity fingerprint may be compared to other entity fingerprints to understand similarities and differences that may exist. In one embodiment, computed entity fingerprints are used to generate feature vectors to be used in classification and clustering tasks.
  • Factors used by the fingerprint extraction module 26 to generate an entity fingerprint relate to the neighborhood of the entity determined from the directed graph.
  • the fingerprint extraction module 26 identifies neighborhood entities of a given type to which the entity is related and temporal aspects of these neighbors (e.g. one attribute being 'emerging neighbor').
  • entity fingerprints computed by the fingerprint-extraction module 26 may represent a point-in-time view of an entity. Structured information about the entity may also be incorporated into the fingerprint.
  • the fingerprint extraction module 26 uses entity to document relationships to add the top-k classification codes and top-k social tags (i.e., Wikipedia article titles related to a document) to organizational fingerprints. This is achieved by sorting each classification code or social tag with respect to the number of documents that include the organization and are also assigned with the classification code or social tag, and selecting the k most frequent results, where k is an integer value.
  • each organizational fingerprint computed by the fingerprint extraction module 26 includes one or more of the following attribute groups: industry hierarchy; geography hierarchy; related industries; related geographies; related macroeconomic indicators; related commodities; related TRCS codes; related social tags; and related entities (in two groupings— those that are emerging and those that are stable.)
  • attribute groups industry hierarchy; geography hierarchy; related industries; related geographies; related macroeconomic indicators; related commodities; related TRCS codes; related social tags; and related entities (in two groupings— those that are emerging and those that are stable.)
  • a graph module 28 is included in the non- volatile memory 20 to construct the directed graph.
  • the graph module 28 initiates construction of the directed graph by selecting documents from available data sources and processing these documents.
  • the processing of these documents includes extracting entities based on metadata associated with a document, and resolving any inconsistencies that may be detected with structured data sources.
  • the graph module 28 then generates vertices of the directed graph based on the extracted entities and generates edges between the extracted entities. Once entities are extracted, the graph module 28 generates edges of the graph between resolved entities.
  • an association module 30 also is provided in the non- olatile memory 20.
  • the association module 30 computes a significance score for an association between entities, the significance score being an indication of the level of significance of one entity to another entity.
  • the association module 30 uses entity cooccurrences in unstructured documents to establish connections (e.g., edges) between entities.
  • the association module 30 may assign a significance score to one or more edges of the directed graph, where higher values indicate stronger connections.
  • entity relationships may be represented using a directed graph, there exists two edges between each pair of connected entities, each of which may be assigned a different significance score by the association module 30.
  • this allows the system 10 to model a common real-life situation where a given entity El may be very significant for a connected entity E2, but the connected entity E2 may not be equally significant for entity El.
  • FIG. 4 An example of entity relationships between one organization, Facebook 100 and some of its entity neighbors is shown in connection with FIG. 4.
  • the example significance scores illustrated in FIG. 4 indicate that Twitter 114 is more significant to Facebook 100, as indicated by a strength of 0.12, than Microsoft 102, indicated by a strength of 0.05, and both are more significant than Apple 104, indicated by a strength of 0,01.
  • Facebook 100 is important to Microsoft 102 but not equally significant, as indicated by strength of 0.01.
  • Microsoft 102 is a major shareholder and investor in Facebook 100, but is also involved in a variety of other business areas.
  • the relationship between Facebook 100 and Google 108 exhibits a similar behavior.
  • association module 30 computes significance scores using a variety of factors that capture different aspects of entity relationships. As noted previously, details of computing significance scores is discussed in U.S. Patent Application No.: 13/107,665 filed May 13, 2011, entitled 'Association Significance', the contents of which is incorporated herein in its entirety.
  • the association module 30 computes significance scores as a weighted average of multiple factors which include interestingness, recent interestingness, validation, common neighbors, industry overlap, geographic overlap, temporal significance, and element of surprise.
  • the interestingness factor is computed by considering entities as two variables, populating a contingency-table with entity frequencies from all available documents, and then applying an interestingness measure to the contingency-table.
  • the recent interestingness factor is computed similar to the interestingness factor, but is computed only using documents that have been created and/or made available over a user- definable time interval.
  • this factor enhances emerging relationships.
  • the validation factor is computed by assigning a value of one (1 ) if the relationship between a source and target entities was validated by a human expert (in available structured data). Otherwise, a value of zero (0) is assigned to the validation factor.
  • the common neighbors factor is computed as a percentage of the degree- 1 neighbors of a source entity that also occur in the degree- 1 neighborhood of the target entity.
  • the industry overlap factor is computed as a percentage of the industries in a source entity's neighborhood that also occur in the degree- 1 neighborhood of the target entity.
  • the geography overlap factor is computed as a percentage of the geographies in source entity's neighborhood that also occur in the degree- 1 neighborhood of the target entity.
  • the temporal significance factor is computed by comparing the recent interestingness value with an interestingness value computed from historic (non-recent) documents.
  • the association module 30 assigns a value of one (1) to the factor if the recent interestingness value is greater than a threshold value, and assigns a value of zero (0) to the factor if the recent interestingness value is less than the threshold value. This factor rewards relationships that had gained strength in the recent time period and penalizes relationships that had lost strength.
  • the element of surprise factor is computed by assigning a value of one (1) if the source entity's neighborhood contains any new industries or geographies in the recent period that did not occur in the historic period, and the target entity shares at least one such industry or geography, otherwise the element of surprise factor is assigned a value of zero (0).
  • association module 30 Additional details of computing significance scores utilized by the association module 30 are disclosed in co-pending U.S. Patent Application No.: 13/107,665 filed May 13, 2011, entitled 'Association Significance', the content of which is incorporated herein in its entirety.
  • the comparison module 32 is provided and computes a similarity score between two fingerprints.
  • the similarity score is computed as a weighted linear combination of the cosine similarity scores of each attribute included in the fingerprints, where weights may be empirically selected by domain experts. This score represents how alike two entities are, rather than simply how connected they are in the graph. This information is of particular utility to investment bankers. A method of comparing entity fingerprints is discussed in connection with FIG. 3,
  • the non- volatile memory 20 also includes an interface module 34 that provides a graphical user interface for viewing none or more entity fingerprints. Additional details of each of the modules 26, 28, 30, 32 and 34 are discussed in further detail below.
  • the software modules 26-34 of the present invention may be implemented as a set of services that are accessible using Flash or HTML-based client applications. Requests and responses may be sent using either standard HTTP, XML over HTTP, or be batched using Google's Protocol Buffers. Data partitioning also may be used to allow vertical and horizontal scalability, i.e., the directed graph may be distributed between multiple nodes in a de-centralized peer network, each of which owns a subset of documents or entities. Protocol Buffers may also be used for inter-node communication to service user requests in the distributed graph.
  • a network 56 is provided that can include various devices such as routers, server, and switching elements connected in an Intranet, Extranet or Internet configuration.
  • the network 56 uses wired communications to transfer information between an access device 58, the server device 12, and a data store 40.
  • the network 56 employs wireless communication protocols to transfer information between the access device 58, the server device 12, and the data store 40.
  • the network 56 employs a combination of wired and wireless technologies to transfer information between the access device 58, the server device 12, and the data store 40.
  • the access device 58 can include a personal computer, laptop computer, or other type of electronic device, such as a cellular phone or Personal Digital Assistant (PDA).
  • the access device 58 is coupled to I/O devices (not shown) that include a keyboard in combination with a pointing device such as a mouse for sending web page requests to the server device 12.
  • I/O devices not shown
  • memory of the access device 58 is configured to include a web browser 58A that is used to request and receive information from the server 12.
  • the system 10 can support multiple access devices.
  • the data store 40 is a repository that maintains and stores information utilized by the before-mentioned modules 26-34.
  • the data store 40 is a relational database.
  • the data store 40 is a directory server, such as a Lightweight Directory Access Protocol ('LDAP').
  • 'LDAP' Lightweight Directory Access Protocol
  • the data store 40 is an area of nonvolatile memory 20 of the server 12.
  • the data store 40 includes a set of entity types 42.
  • entity types 42 As used herein, the words 'set' and 'sets' refer to anything from a null set to a multiple element set.
  • Each member of the set of entity types 42 represents a type of tangible or logical entity that is pre-defined in the system.
  • the set of entity types 42 may include, but are not limited to, the following classes of entities: organizations, people, products, industries, geographies, commodities, financial indicators, economic indicators, events, topics, subject codes, unique identifiers, social tags, industry terms, general terms, metadata elements, classification codes, and combinations thereof.
  • a set of pre-defined entities 44 are also provided that are used to identify similar entities and to determine attributes of entities.
  • Each member of the set of pre-defined entities represents a specific tangible or logical entity and is associated with zero (0), one (1), or many members of the set of entity types 42.
  • each member of the set of pre-defined entities 44 may be associated with an attribute identifier, an attribute-type identifier, and an attribute value (i.e., a triplet). For example, if one of the pre-defined entities represents a company, such as 'Thomson Reuters', an attribute-type identifier may be 'Corporate headquarters', an attribute identifier may be 'City", and an attribute value may be "New York".
  • the data store 40 includes a set of information sources 46.
  • the set of information sources 46 are structured data sources in which each member is associated with one (1) or more attribute identifiers, attribute-type identifiers, attribute values, and information source identifiers (i.e., a quadruple) for each member of the set of pre-defined entities 44.
  • the data store 40 also includes a set of documents 48.
  • Each of the set of documents 48 is optionally associated with a set of document quadruples comprising an entity identifier (e), an attribute identifier, an attribute-type identifier, and an attribute value.
  • entity identifier (e) relates to at least one member of the set of pre-defined entities 44.
  • at least one document of the set of documents is associated with metadata that identifies one or more entities.
  • a set of weights 50 also is provided in the data store 40 and is used to compute similarities between fingerprint attributes.
  • Each member of the set of weights 50 is associated with a numerical value and one or more information source identifier, attribute-type identifier, and binary variables indicating whether the numerical weight value is obtained from the directed graph and whether the numerical weight value has been validated by a human.
  • a set of association criteria 52 is provided that comprises contingency tables used by the association module 30 to compute a significance score for an identified relationship between entities.
  • the contingency tables are associated with a set of interestingness measures that are used by the association module 30 to compute the significance score.
  • the data store 40 also includes a graph object 54.
  • the graph object 54 is a serialized Java object that has been persisted and represents the directed graph. Information needed to generate the directed graph is stored in one or more data sources stored in the data store 40. Additional details concerning information included in the data store 40 are discussed in greater detail below.
  • data store 40 shown in FIG. 1 is connected to the network 56, it will be appreciated by one skilled in the art that the data store 40 and/or any of the information shown therein, can be distributed across various servers and be accessible to the server 12 over the network 56, be coupled directly to the server 12, or be configured in an area of non- volatile memory 20 of the server 12.
  • system 10 shown in FIG. 1 is only one embodiment of the disclosure.
  • Other system embodiments of the disclosure may include additional structures that are not shown, such as secondary storage and additional computational devices.
  • various other embodiments of the disclosure include fewer stractures than those shown in FIG. 1.
  • the disclosure is implemented on a single computing device in a non-networked standalone configuration. Data input and requests are communicated to the computing device via an input device, such as a keyboard and/or mouse. Data output, such as the computed significance score, of the system is communicated from the computing device to a display device, such as a computer monitor.
  • the fingerprint extraction module 26 initializes a fingerprint 'F' for the entity ⁇ * .
  • the fingerprint A comprises a data structure having at least one attribute identifier that identifies an entity characteristic, an attribute- type identifier that identifies a category for the entity
  • the data structure comprises zero and/or null values.
  • the fingerprint extraction module 26 compares entity A with each member of the set of pre-defined entities 44. Based on the comparison, the fingerprint extraction module 26 adds a quadruple to fingerprint F for each identified member in the set of pre-defined entities 44 by importing values associated with identified entities into the data structure. The fingerprint extraction module 26 then determines the attribute weight of the quadruple from the weights data store 50 using the attribute-type identifier and the information source identifier from the set of information sources 46.
  • the fingerprint extraction module 26 identifies all members of the set of information sources 46 that comprise a quadruple for entity A.
  • the fingerprint extraction module 26 adds a quadruple to the fingerprint F for each identified member of the set of information sources 46 by importing attribute identifiers, attribute-type identifiers, attribute values, and information source identifiers directly from the set of information sources 46.
  • the fingerprint extraction module 26 determines the attribute weight from the weight data store 50 using the attribute-type identifier and an information source identifier associated with each identified member of the set of information sources 46.
  • the fingerprint extraction module 26 adds a quadruple to fingerprint F for each vertex that meets user-defined criteria.
  • the user-defined criteria may be used to identify vertices connected via an edge that have a significance or relevance score attribute that meets a user-defined threshold value.
  • the user-defined criteria may also be used to identify vertices associated with a user- defined attribute- type identifier or attribute identifier, and/or edges that are associated with a user-defined attribute-type identifier or attribute identifier.
  • the fingerprint extraction module 26 then imports values for the fingerprint A from each identified vertex (e.g., attribute identifiers, attribute-type identifiers, and attribute values associated with each vertex) and determines attribute weights from the weights data store 50 using the attribute-type, and binary values indicating that the attribute is derived from the directed graph and not the set of information sources 46.
  • values for the fingerprint A e.g., attribute identifiers, attribute-type identifiers, and attribute values associated with each vertex
  • the fingerprint extraction module 26 processes a set of edges that directly connect to entity A to obtain attribute metrics of connected entities.
  • the fingerprint extraction module 26 adds a quadruple to fingerprint F using a calculated attribute identifier and value that is based on an identified attribute-type.
  • the fingerprint extraction module 26 determines an attribute weight for the calculated attribute from the weights data store 50 using the attribute-type, and binary values indicating that the attribute is derived from the directed graph and not the set of information sources 46.
  • the fingerprint extraction module 26 imports the calculated attribute identifier, attribute-type, attribute value, and attribute weight into the fingerprint F data structure.
  • at least one document of the set of documents 48 may include metadata tags and/or entity information that identifies one or more entities.
  • the fingerprint extraction module 26 counts the number of documents that exclude entity A and adds a quadruple to fingerprint F for each document, selecting entities that meet a user-defined criteria, such as a threshold value indicative of a minimum number of documents that contain the selected entities.
  • the fingerprint extraction module 26 imports values for the fingerprint data structure from identified documents based on the ranking of metadata tags and/or entity information, and at step 72, stores references to the identified documents in the directed graph (e.g., vertices and edges of the directed graph) based on the ranking.
  • the comparison module 32 identifies a first entity associated with a first fingerprint and a second entity associated with a second fingerprint.
  • the first entity and the second entity are initially identified by a user and received via a graphical user interface provided by the interface module 34 using unique identifiers associated with each entity.
  • the comparison module 32 computes a first weighted value for each unique attribute-type associated with the first entity.
  • the comparison module 32 applies an aggregation function, such as an average function, a minimum function, or maximum function, to attribute weights of attribute identifiers that are associated with the first entity.
  • the comparison module 32 computes a second weiglited value for each unique attribute-type associated with the second entity. Similar to computing the first weighted value, the comparison module 32 applies an aggregation function to attribute weights of attribute identifiers associated with the second entity.
  • the comparison module 32 normalizes attribute values of the first entity using the computed first weighted value of the first entity, and normalizes attribute values of the second entity using the computed second weighted value of the second entity.
  • Each of these computed weighted values are equally distributed across each respective entity's attribute values. For example, assuming a computed weighted value of twenty (20) for an attribute-type TY and a number of attribute identifiers for the attribute-type TY of ten (10), the comparison module 32 assigns an attribute weight of two (2) (i.e., aggregated weight(20)/number of attributes(lO)) to each attribute corresponding to the attribute-type TY.
  • the comparison module 32 computes a similarity score for the first and second entity fingerprints by applying similarity functions to corresponding normalized attributes of each fingerprint and aggregates the results.
  • Example similarity functions that may be applied to corresponding normalized attributes include, but are not limited to, cosine similarity,
  • the comparison module 32 generates a signal associated with the first entity and the second entity and transmits the signal in response to a request.
  • GUI 140 includes a query frame 142 for specifying an entity, a matrix frame 144 for listing similar entity fingerprints to the entity specified in the query frame 142, and a headline frame 146 for listing any news items and/or documents that may relate to the entity specified in the query frame 142.
  • an input text field 142A is provided that allows a user to specify a unique entity identifier.
  • the unique entity identifier may be, but is not limited to, a ticker symbol, entity code, entity name, or partial entity name.
  • the unique entity identifier is associated with entities modeled in the system, in one embodiment, the input text field 142 A is auto suggest enabled for entities of type 'Company'. Upon selecting a magnifying glass or pressing an enter key, the unique entity identifier is received by the fingerprint extraction module 26.
  • the fingerprint extraction module 26 then formulates a query for the entity and returns the entity fingerprint and similar fingerprints for display in the query frame 142, the matrix frame 144, the headline frame 146.
  • an entity fingerprint is a predefined set of characteristics/entities that reflect important aspects of the entity. For example, in the case of an entity of type 'Company', sector membership, related commodities and geographic information may be associated with the entity fingerprint.
  • a user is allowed to temporarily remove
  • the query frame 142 is configured to include a trend line 142B for the selected query entity (i.e., the entity represented by the entity identifier in the input text field 142A).
  • the trend line 142B is displayed as a chart with a thirty (30) day history. Each day is represented as a bar and represents a number of documents mentioning the query entity.
  • the matrix frame 144 displays entities that have been determined by the fingerprint extraction module 26 to have the most shared entities/attributes with the query entity.
  • the similar entities are ranked by the fingerprint extraction module 26 and then displayed in a list with the most similar entity to the query entity being displayed at the top of the list.
  • the list is provided with a scroll capability to display multiple similar entities.
  • each similar entity may be dragged and dropped to the input field 142A of the query frame 142. This operates to replace the original query entity with the dragged entity and update all frames 142, 144, 146 accordingly.
  • each identified similar entity 152, 154, 156 is displayed with its entity name, percentage of attributes in common with the query entity 158, percentage of shared connections with the query entity from the directed graph 160, a number of top connections not shared with the query entity 162, and whether there are any emerging or trending connections for the similar entityl64.
  • a selectable trend line 166 associated with each similar entity illustrates the number of documents noting the similar entity, per day, for a preset time interval (e.g., the last thirty (30) days).
  • a large overlay 170 of the trend line 166 is displayed.
  • the overlay 170 allows a user to drill down to specific news events for the entity.
  • a pointing device such as a mouse
  • any bar of the overlay 170 results in a call out 172 being displayed that includes individual headlines relating to the similar entity.
  • FIG. 8 illustrates the matrix frame 144 having a horizontal grid line 180.
  • a user is able expand the horizontal grid line 180 in the matrix frame 144 by clicking on an arrow 180A next to a value. As shown in the FIG. 8 example, this operates to push displayed results below the grid line 180 resulting in more detailed information on reasons entities have been found similar to each other being presented to the user.
  • the headline frame 146 displays a selectable list 146 A of the most important headlines/ documents for the query entity, as well as all the similar entities displayed in the matrix frame 144, over a predefined time interval.
  • the selectable list 146A is sorted by time, with the most recent headline/document displayed at the top of the list.
  • a pointing device such as a mouse
  • a popup dialog box is displayed listing the headline, a date, and one or more entities the headline/document references.
  • an entity fingerprint display 150 having a plurality of adjacent navigation options for accessing an adjacent view of entities.
  • the adjacent view allows a user to view common and uncommon entities and concepts shared between two entities.
  • the adjacent view may be accessed via a push button 154 included in a top navigation menu 152 of the fingerprint display 150 and also via a navigation item 156 associated with each result item of the matrix frame 144
  • FIG. 12 is a schematic of a side-by-side display 160. Referring now to both FIGS.
  • any query entity currently displayed in the input text field 142A of the fingerprint display 150 is passed to the side-by-side display 160.
  • the passed query entity is displayed on the left side (area 1) 162 of the side-by-side display 160.
  • the selected entity identifier is sent to the side-by-side display 160 and displayed on the right side (area 2) 164 of the side-by-side display. If the user has not selected any of the items in the matri frame 144, the right side (area 2) 164 of the side-by-side display 160 remains unpopulated.
  • the query entity displayed in the input text field 142 A of the fingerprint display 150 is passed to the side-by-side display 160 and displayed on the left side (area 1) 162 of the side-by- side display 160,
  • the selected result Sine item is passed to the side-by-side display and displayed on the right side (area 2) 164,
  • Result area 166 is used to display connection strengths between entities and is discussed in further detail below.
  • FIG. 13 an example side-by-side display 170 showing a first entity identifier 172 ('Apple Inc.') and a second entity identifier 174 ('Intel Corp.') is shown.
  • the comparison module 32 forms a query of the two entities, hereinafter referred to as the 'query entities', and populates the result area 166 with a visual depiction of entity connection strengths between the query entities.
  • the result area 166 comprises a number of horizontal lines 178.
  • Each horizontal line 178 represents an entity type, such as but not limited to TRCS sectors, companies, places, commodities, geographies, people, corporate events, and economic indicators, and is populated by the comparison module 32 with a number of entities of the same entity type.
  • each entity displayed on a horizontal line is represented using a visual indicator, such as a ball.
  • the query entities operate as 'magnets.' Specifically, if one of the visual indicators representing an entity is primarily related to one of the query entities, the visual indicator is displayed closer to that query entity's side of the result area 166. Entities with the same or similar strength of connections to both entities are displayed toward the middle zone of the result area 166.
  • the size of the visual indicator may be determined based on the relative importance of the entity to the query entities combined. For example, in one embodiment, referring to the FIG. 13 example, a larger ball depicted in the result area 166 indicates that the entity depicted by the larger ball is of greater importance to the query entities than a ball shown of a smaller dimension.
  • the visual identifiers are associated with labels that are automatically displayed.
  • a label associated with the selected visual identifier is displayed to the user.
  • visual identifiers may be color coded providing different visual aspects of entities being displayed.
  • Various features of the system may be implemented in hardware, software, or a combination of hardware and software.
  • some features of the system may be implemented in one or more computer programs executing on programmable computers.
  • Each program may be implemented in a high level procedural or object-oriented programming language to communicate with a computer system or other machine,
  • each such computer program may be stored on a storage medium such as read-only-memory (ROM) readable by a general or special purpose programmable computer or processor, for configuring and operating the computer to perform the functions described above.
  • ROM read-only-memory

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

L'invention concerne des systèmes et techniques pour l'exploration de relations parmi des entités. Les systèmes et techniques fournissent une analyse d'informations à base d'entité et une plate-forme de groupage de contenu qui utilise des sources de données hétérogènes pour construire et maintenir un écosystème autour d'entités tangibles et logiques. Les entités sont représentées comme des sommets dans un graphe orienté, et les bords sont générés à l'aide de co-occurrences d'entités dans des documents non structurés et des informations supervisées provenant de sources de données structurées. Des résultats d'importance pour les bords sont calculés à l'aide d'un procédé qui combine des facteurs supervisés, non supervisés et temporels en un résultat unique, des attributs d'entités importants provenant du contenu structuré et du voisinage d'entités dans le graphe sont automatiquement analysés comme l'empreinte digitale d'entité. Les entités peuvent être comparées les unes aux autres d'après la similitude de leurs empreintes digitales d'entités. L'invention concerne également une interface utilisateur interactive qui fournit un accès exploratoire au graphe et accepte des processus d'aide à la décision.
PCT/US2012/025942 2011-02-22 2012-02-21 Empreintes digitales d'entités Ceased WO2012115965A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201280019646.3A CN103827895B (zh) 2011-02-22 2012-02-21 实体指纹
EP12707690.9A EP2678809A1 (fr) 2011-02-22 2012-02-21 Empreintes digitales d'entités

Applications Claiming Priority (10)

Application Number Priority Date Filing Date Title
US201161445236P 2011-02-22 2011-02-22
US61/445,236 2011-02-22
US201161486091P 2011-05-13 2011-05-13
US13/107,665 US9495635B2 (en) 2011-02-22 2011-05-13 Association significance
US61/486,091 2011-05-13
US13/107,665 2011-05-13
US201161519592P 2011-05-25 2011-05-25
US61/519,592 2011-05-25
US13/213,324 US9292545B2 (en) 2011-02-22 2011-08-19 Entity fingerprints
US13/213,324 2011-08-19

Publications (1)

Publication Number Publication Date
WO2012115965A1 true WO2012115965A1 (fr) 2012-08-30

Family

ID=46721196

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2012/025942 Ceased WO2012115965A1 (fr) 2011-02-22 2012-02-21 Empreintes digitales d'entités

Country Status (3)

Country Link
EP (1) EP2678809A1 (fr)
CN (1) CN103827895B (fr)
WO (1) WO2012115965A1 (fr)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2014182864A3 (fr) * 2013-05-09 2015-01-22 Microsoft Corporation Déduction de valeurs d'attributs d'entités
US20170286565A1 (en) * 2011-06-23 2017-10-05 FullContact, Inc. Relationship graph
US10474702B1 (en) 2014-08-18 2019-11-12 Street Diligence, Inc. Computer-implemented apparatus and method for providing information concerning a financial instrument
US11144994B1 (en) 2014-08-18 2021-10-12 Street Diligence, Inc. Computer-implemented apparatus and method for providing information concerning a financial instrument
US11574287B2 (en) 2017-10-10 2023-02-07 Text IQ, Inc. Automatic document classification

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016003512A1 (fr) * 2014-06-30 2016-01-07 Linkedin Corporation Détermination d'un type de relation entre entités disparates
US10523736B2 (en) 2014-06-30 2019-12-31 Microsoft Technology Licensing, Llc Determining an entity's hierarchical relationship via a social graph
US9424458B1 (en) * 2015-02-06 2016-08-23 Hoyos Labs Ip Ltd. Systems and methods for performing fingerprint based user authentication using imagery captured using mobile devices
US11263432B2 (en) 2015-02-06 2022-03-01 Veridium Ip Limited Systems and methods for performing fingerprint based user authentication using imagery captured using mobile devices

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060282415A1 (en) * 2005-06-09 2006-12-14 Fuji Xerox Co., Ltd. Document retrieval apparatus

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101345691A (zh) * 2007-07-10 2009-01-14 朱东华 数据分析报告智能化生成并通过无线有线技术收发的方法
US8504568B2 (en) * 2009-01-08 2013-08-06 Fluid Operations Gmbh Collaborative workbench for managing data from heterogeneous sources
US20100318512A1 (en) * 2009-06-16 2010-12-16 Ludwig Lester F Advanced geographic information system (gis) providing modeling, decision support, visualization, sonification, web interface, risk management, sensitivity analysis, sensor telemetry, field video, and field audio
CN101950302B (zh) * 2010-09-29 2011-12-21 李晓耕 基于移动设备的海量音乐库管理方法

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060282415A1 (en) * 2005-06-09 2006-12-14 Fuji Xerox Co., Ltd. Document retrieval apparatus

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
COOPER J W ET AL: "Detecting Similar Documents Using Salient Terms", 4 November 2002 (2002-11-04), pages 1 - 6, XP002587228, ISBN: 978-1-58113-492-6, Retrieved from the Internet <URL:http://portal.acm.org/ft_gateway.cfm?id=584835&type=pdf&coll=GUIDE&dl=GUIDE&CFID=95314629&CFTOKEN=91612696> [retrieved on 20120717] *
HASSAN H. MALIK ET AL: "Exploring the corporate ecosystem with a semi-supervised entity graph", PROCEEDINGS OF THE 20TH ACM CONFERENCE ON INFORMATION AND KNOWLEDGE MANAGEMENT, CIKM 201, 28 October 2011 (2011-10-28), pages 1857 - 1866, XP055032624, ISBN: 978-1-45-030717-8, Retrieved from the Internet <URL:http://www.cs.columbia.edu/~hhm2104/papers/Atlas_CIKM_2011.pdf> [retrieved on 20120712] *
HENNING RODE: "From Document to Entity Retrieval", CTIT PH.D. THESIS SERIES NO.08-120, 31 December 2008 (2008-12-31), University Twente, Netherlands, XP055032920, Retrieved from the Internet <URL:http://doc.utwente.nl/60765/1/thesis_H_Rode.pdf> [retrieved on 20120717] *
TORU TAKAKI ET AL: "Associative document retrieval by query subtopic analysis and its application to invalidity patent search", PROCEEDINGS OF THE THIRTEENTH ACM CONFERENCE ON INFORMATION AND KNOWLEDGE MANAGEMENT , CIKM '04, 1 January 2004 (2004-01-01), New York, New York, USA, pages 399, XP055032925, DOI: 10.1145/1031171.1031251 *

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170286565A1 (en) * 2011-06-23 2017-10-05 FullContact, Inc. Relationship graph
US10445371B2 (en) * 2011-06-23 2019-10-15 FullContact, Inc. Relationship graph
US11216508B2 (en) 2011-06-23 2022-01-04 FullContact, Inc. Entity resolution
WO2014182864A3 (fr) * 2013-05-09 2015-01-22 Microsoft Corporation Déduction de valeurs d'attributs d'entités
US9501503B2 (en) 2013-05-09 2016-11-22 Microsoft Technology Licensing, Llc Inferring entity attribute values
US10394854B2 (en) 2013-05-09 2019-08-27 Microsoft Technology Licensing, Llc Inferring entity attribute values
US10474702B1 (en) 2014-08-18 2019-11-12 Street Diligence, Inc. Computer-implemented apparatus and method for providing information concerning a financial instrument
US11144994B1 (en) 2014-08-18 2021-10-12 Street Diligence, Inc. Computer-implemented apparatus and method for providing information concerning a financial instrument
US11574287B2 (en) 2017-10-10 2023-02-07 Text IQ, Inc. Automatic document classification
US12125000B2 (en) 2017-10-10 2024-10-22 Text IQ, Inc. Automatic document classification

Also Published As

Publication number Publication date
CN103827895B (zh) 2017-08-08
EP2678809A1 (fr) 2014-01-01
CN103827895A (zh) 2014-05-28

Similar Documents

Publication Publication Date Title
US10235425B2 (en) Entity fingerprints
US12124492B2 (en) User interface for visualizing search data
US11868411B1 (en) Techniques for compiling and presenting query results
EP2678809A1 (fr) Empreintes digitales d&#39;entités
US11328307B2 (en) Brand abuse monitoring system with infringement detection engine and graphical user interface
US11429405B2 (en) Method and apparatus for providing personalized self-help experience
US8635281B2 (en) System and method for attentive clustering and analytics
Lahby et al. Online fake news detection using machine learning techniques: A systematic mapping study
US11250065B2 (en) Predicting and recommending relevant datasets in complex environments
US20110264651A1 (en) Large scale entity-specific resource classification
US20200301973A1 (en) Personalization Aggregate Content Item Recommendations
KR20120049180A (ko) 인공지능 보조 메디컬 레퍼런스 시스템 및 방법
Adams et al. Thematic signatures for cleansing and enriching place-related linked data
García Peñalvo et al. Discovering knowledge through highly interactive information based systems
US20150302097A1 (en) System for classifying characterized information
CN113762703A (zh) 确定企业画像的方法和装置、计算设备和存储介质
CA2956627A1 (fr) Systeme et moteur servant au regroupement cible d&#39;evenements d&#39;informations
CN114398864A (zh) 报表显示方法、装置、设备及存储介质
Wu et al. Updating high average-utility itemsets with pre-large concept
Ye et al. An interpretable mechanism for personalized recommendation based on cross feature
Shringi et al. Detection of spam reviews using hybrid grey wolf optimizer clustering method
CN114912966B (zh) 信息推送方法、装置、电子设备和存储介质
Solainayagi et al. Trust discovery and information retrieval using artificial intelligence tools from multiple conflicting sources of web cloud computing and e-commerce users
WO2016203230A1 (fr) Procédé et système de localisation d&#39;informations de réglementation
CN114091860A (zh) 查询方法、装置、计算机设备、存储介质和程序产品

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 12707690

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

WWE Wipo information: entry into national phase

Ref document number: 2012707690

Country of ref document: EP