Generating a domain-specific knowledge graph from unstructured computer text

Patent metadata

Field	Value
Publication number	US-12541694-B2
Application number	US-202017105131-A
Country	US
Kind code	B2
Filing date	Nov 25, 2020
Priority date	Nov 25, 2020
Publication date	Feb 3, 2026
Grant date	Feb 3, 2026

Abstract

Methods and apparatuses are described for generating a domain-specific knowledge graph from unstructured computer text. A computing device extracts unstructured computer text associated with pairs of entities from domain-independent documents, and trains an entity relationship classification model using a domain-independent knowledge graph and the extracted text. The computing device extracts other unstructured text associated with a first domain from domain-specific documents. The computing device identifies pairs of entities contained within the text for the first domain, and executes the trained model to determine a relationship between the entities in each pair of entities identified from the text for the first domain. The computing device generates a domain-specific knowledge graph using (i) the pairs of entities identified from the text for the first domain and (ii) the relationships between the entities in each pair of entities identified from the text for the first domain.

First claim

What is claimed is:

1 . A system for generating a domain-specific knowledge graph from unstructured computer text for use in a conversation service application, the system comprising:
a computer data store including (i) a plurality of domain-specific documents, each document comprising unstructured computer text associated with a first domain (ii) a domain-independent knowledge graph comprising a plurality of pairs of entities, each pair including a relationship between the entities in the pair, and (iii) a plurality of domain-independent documents, each document comprising unstructured computer text associated with one or more of the plurality of pairs of entities; and a computing device coupled to the computer data store, the computing device comprising a memory that stores computer-executable instructions and a processor that executes the computer-executable instructions to:
extract the unstructured computer text associated with one or more of the plurality of pairs of entities from at least a portion of the plurality of domain-independent documents;
train an entity relationship classification model using the domain-independent knowledge graph and the unstructured computer text extracted from the at least a portion of the plurality of domain-independent documents, the entity relationship classification model configured to identify entity relationships contained in input data received by the model,
wherein the entity relationship classification model comprises an encoder and a decoder, and training includes encoding, at the encoder using a piecewise convolutional neural network, segmentations of sentences of the unstructured computer text as segmented by the associated pair of entities and assigning, at the decoder, each encoded sentence of the unstructured computer text to an entity pair bag and predicting a relationship classification for each entity pair bag;

extract the unstructured computer text associated with the first domain from at least a portion of the plurality of domain-specific documents and combine the extracted unstructured computer text into a corpus of unstructured computer text for the first domain;
identify, using a named entity recognition model, a plurality of pairs of entities contained within the corpus of unstructured computer text for the first domain, each pair of entities located in one or more sentences of the corpus of unstructured computer text;
execute the trained entity relationship classification model, with input of (i) the plurality of pairs of entities identified from the corpus of unstructured computer text for the first domain and (ii) the sentences for each of the plurality of pairs of entities identified from the corpus of unstructured computer text for the first domain, to determine a relationship between the entities in each pair of the plurality of pairs of entities identified from the corpus of unstructured computer text for the first domain;
generate a domain-specific knowledge graph using (i) the plurality of pairs of entities identified from the corpus of unstructured computer text for the first domain and (ii) the relationship between the entities in each pair of the plurality of pairs of entities identified from the corpus of unstructured computer text for the first domain,
the domain-specific knowledge graph comprising a multidimensional data structure where:
each entity in the plurality of pairs of entities identified from the corpus of unstructured computer text for the first domain is a node in the domain-specific knowledge graph, and
the relationship between the entities in each pair the plurality of pairs of entities identified from the corpus of unstructured computer text for the first domain is a connection between the nodes for those entities in the domain-specific knowledge graph;

establish a chat-based communication session between a conversation service software application of the computing device and a remote computing device;
capture a user message generated by a user of the remote computing device during the chat-based communication session;
traverse the domain-specific knowledge graph using one or more keywords extracted from the first message to identify a first data element in the knowledge graph that is responsive to the user message and one or more second data elements in the knowledge graph that are related to the first data element;
generate (i) a first responsive message containing the first data element and (ii) a second responsive message containing the second data elements, wherein at least one of the first data element or the second data elements comprises a hyperlink that points to a web address containing digital content; and
transmit the first responsive message and the second responsive message to the remote computing device for display to the user during the chat-based communication session,
wherein the hyperlink is activated at the remote computing device to establish a connection to a web server at the web address for retrieval and display of the digital content on the remote computing device alongside the conversation service software application.

2 . The system of claim 1 , wherein extracting the unstructured computer text associated with one or more of the plurality of pairs of entities from at least a portion of the plurality of domain-independent documents comprises segmenting each document in the at least a portion of the plurality of domain-independent documents into a plurality of sentences.

3 . The system of claim 2 , wherein training the entity relationship classification model using the domain-independent knowledge graph and the unstructured computer text extracted from the at least a portion of the plurality of domain-independent documents comprises:
selecting seed entities from the domain-independent knowledge graph; identifying a plurality of pairs of entities from the domain-independent knowledge graph using the seed entities; capturing sentences from the plurality of domain-independent documents that are associated with the identified plurality of pairs of entities from the domain-independent knowledge graph; determining a relationship between entities in each pair of the identified plurality of pairs of entities from the domain-independent knowledge graph based upon the captured sentence from the plurality of domain-independent documents that is associated with the pair; and training the entity relationship classification model using as input (i) the identified plurality of pairs of entities from the domain-independent knowledge graph, (ii) the determined relationships in each pair of the identified plurality of pairs of entities from the domain-independent knowledge graph, and (iii) captured sentences from the plurality of domain-independent documents that are associated with the identified plurality of pairs of entities from the domain-independent knowledge graph.

4 . The system of claim 3 , wherein capturing sentences from the plurality of domain-independent documents is based upon a distant supervision algorithm.

5 . The system of claim 1 , wherein extracting the unstructured computer text associated with the first domain from at least a portion of the plurality of domain-specific documents comprises segmenting each document in the at least a portion of the plurality of domain-specific documents into a plurality of sentences.

6 . The system of claim 1 , wherein identifying, using the named entity recognition model, a plurality of pairs of entities contained within the corpus of unstructured computer text for the first domain comprises:
converting, by a word embedding layer of the named entity recognition model, each word of the sentences of the corpus of unstructured computer text for the first domain into a multidimensional vector in a multidimensional vector spa

Assignees

• Fmr Llc

Inventors

• Hao Hua
• Guo Tieyi
• Chun Byung
• Ren Wenbin
• Yu Yan

Classifications

• H04L51/02

  using automatic reactions or user delegation, e.g. automatic replies or chatbot-generated messages · CPC title

• G06N5/02

  Knowledge representation; Symbolic representation · CPC title

• G06N3/08

  Learning methods · CPC title

• G06F40/295 Primary

  Named entity recognition · CPC title

• G06F16/93

  Document management systems · CPC title