Back to results
US 10,706,113 B2Granted
Domain review system for identifying entity relationships and corresponding insights
AssigneeMicrosoft Technology Licensing, LLC
InventorsJessica Lundin, Ryen W. White, Kris K. Ganjam, Navendu Jain, Hua He
Life of this patent
Estimating the term…
Abstract
A system and method is provided for generating a dynamic comprehensive domain review. A domain review engine obtains authoritative literature associated with a domain to extract insights using entity recognition and relationship extraction, and ranks the extracted results to generate a dynamic domain review.
Claims20
A system for generating a dynamic comprehensive domain review, the system comprising:
- a memory;
- at least one processor communicatively coupled to the memory;
- a long short-term memory model network, implemented on the at least one processor, trained to:
- receive authoritative literature as training data,
- identify one or more domains associated with the received authoritative literature,
- recognize conclusory content of the authoritative literature, and
- extract a plurality of entities from the conclusory content;
- and
- a domain search engine, implemented on the at least one processor, that:
- receives a request to generate the dynamic comprehensive domain review,
- identifies a particular domain and at least one contextual parameter associated with the received request, and
- generates the dynamic comprehensive domain review based at least on the trained long short-term memory model network, by implementing the at least one processor to:
- obtain literature associated with the identified particular domain,
- perform entity recognition on text parsed from the literature to identify one or more entities,
- determine one or more relationships between the identified one or more entities at least by implementing the at least one processor to:
- apply a causal relationship function to at least a pair of the identified one or more entities in the long short-term memory model network, and
- generate a vector representation of a causal function that encodes semantic association with the plurality of extracted entities in the long short-term memory model network such that a relative proximity of position in the vector representation in the long short-term memory model network indicates a common context among the pair of the identified one or more entities,
- extract a plurality of insights from the obtained literature based on the one or more determined relationships, and
- rank the extracted plurality of insights based at least in part on analysis of the determined one or more relationships and at least in part on the identified at least one contextual parameter associated with the received request.
16 more elements
A method for generating a comprehensive domain review, the method comprising:
- training a long short-term memory model network comprising:
- receiving authoritative literature as training data,
- identifying one or more domains associated with the received authoritative literature,
- recognizing conclusory content of the authoritative literature,
- extracting a plurality of entities from the conclusory content, receiving a request to generate the comprehensive domain review;
- identifying a particular domain and at least one contextual parameter associated with the received request;
- and
- generating the comprehensive domain review based at least on the trained long short-term memory model network, the generating comprising:
- obtaining, by a domain review module implemented on at least one processor, a body of literature associated with the identified particular domain,
- performing entity recognition on text parsed from the body of literature to identify one or more entities,
- determining one or more relationships between the identified one or more entities at least by:
- applying a causal relationship function to at least a pair of the identified one or more entities in the long short-term memory model network, and
- generating a vector representation of a causal function that encodes semantic association with the plurality of extracted entities in the long short-term memory model network such that a relative proximity of position in the vector representation in the long short-term memory model network indicates a common context among the pair of the identified one or more entities,
- extracting one or more insights from the body of literature based on the one or more determined relationships, and
- ranking the one or more extracted insights based at least in part on analysis of the determined one or more relationships and at least in part on the identified at least one contextual parameter associated with the received trigger.
12 more elements
One or more computer storage devices having computer-executable instructions stored thereon for providing comprehensive domain review, which, on execution by a computer, cause the computer to perform operations comprising:
- training a long short-term memory model network comprising:
- receiving authoritative literature as training data,
- identifying one or more domains associated with the received authoritative literature,
- recognizing conclusory content of the authoritative literature,
- extracting a plurality of entities from the conclusory content,
- responsive to receiving a request to generate the comprehensive domain review, identifying a particular domain associated with the received request;
- generating the comprehensive domain review based at least on the trained long short-term memory model network, the generating comprising:
- obtaining, by a domain review module, a body of literature associated with the identified particular domain,
- performing entity recognition on text parsed from the body of literature to identify one or more entities,
- determining one or more relationships between the one or more identified entities, wherein the determining comprises:
- applying a causal relationship function to at least a pair of the identified one or more entities in the long short-term memory model network, and
- generating a vector representation of a causal function that encodes semantic association with in the long short-term memory model network such that a relative proximity of position in the vector representation in the long short-term memory model network indicates a common context among the pair of the identified one or more entities;
- extracting one or more insights from the body of literature based on the one or more determined relationships;
- and
- ranking the one or more extracted insights based at least in part on confidence scores corresponding to the one or more determined relationships and at least in part on the identified at least one contextual parameter associated with the received trigger.
12 more elements
Dependencies are inferred from the claim text.