Unsupervised event extraction

Patent metadata

Field	Value
Publication number	US-2021365306-A1
Application number	US-202016880556-A
Country	US
Kind code	A1
Filing date	May 21, 2020
Priority date	May 21, 2020
Publication date	Nov 25, 2021
Grant date	—

Abstract

Computer-implemented techniques for unsupervised event extraction are provided. In one instance, a computer implemented method can include parsing, by a system operatively coupled to a processor, unstructured text comprising event information to identify candidate event components. The computer implemented method can further include employing, by the system, one or more unsupervised machine learning techniques to generate structured event information defining events represented in the unstructured text based on the candidate event components.

First claim

What is claimed is:

1 . A system, comprising:
a memory that stores computer executable components; a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise:
a parsing component that parses unstructured text comprising event information to identify candidate event components; and
an event extraction component that generates structured event information defining events represented in the unstructured text based on the candidate event components.

2 . The system of claim 1 , wherein the event extraction component employs one or more unsupervised machine learning techniques to generate the structured event information based on the candidate event components.

3 . The system of claim 1 , wherein the candidate event components comprise one or more candidate event trigger terms and one or more candidate event arguments respectively associated with the one or more candidate event trigger terms.

4 . The system of claim 1 , wherein the computer executable components further comprise:
an event representation component that generates one or more event representations based on the candidate event components.

5 . The system of claim 4 , wherein the event representation component employs graph embeddings to generate the one or more event representations.

6 . The system of claim 4 , wherein the computer executable components further comprise:
a clustering component that employs the one or more event representations to cluster the candidate event components into different event types.

7 . The system of claim 6 , wherein the event extraction component generates the structured event information based on the different event types and the candidate event components respectively grouped with the different event types.

8 . The system of claim 6 , wherein the candidate event components comprise candidate event trigger terms and candidate event arguments respectively associated with the candidate event trigger terms, and wherein the computer executable components further comprise:
a role labeling component that labels the candidate event arguments with one or more role attributes representative of one or more roles the candidate event arguments play with respect to the different event types, wherein the event extraction component generates the structured event information based on the different event types, the candidate event trigger terms respectively associated with the different event types, the candidate event arguments respectively associated with the candidate event trigger terms, and the one or more role attributes.

9 . The system of claim 8 , wherein the role labeling component employs one or more external knowledge bases to facilitate labeling the candidate event arguments with the one or more role attributes.

10 . The system of claim 1 , wherein the parsing component employs abstract meaning representation parsing to identify the candidate event components.

11 . The system of claim 1 , wherein the unstructured text comprises candidate event components and relevant event information included in one or more unstructured text documents.

12 . The system of claim 1 , wherein the computer executable components further comprise:
a query component that receives a query request regarding an event and employs the structured event information to identify one or more parts of the unstructured text that are relevant to the query request.

13 . A method, comprising:
parsing, by a system operatively coupled to a processor, unstructured text comprising event information to identify candidate event components; and employing, by the system, one or more unsupervised machine learning techniques to generate structured event information defining events represented in the unstructured text based on the candidate event components.

14 . The method of claim 13 , wherein the candidate event components comprise one or more candidate event trigger terms and one or more candidate event arguments respectively associated with the one or more candidate event trigger terms.

15 . The method of claim 13 , further comprising:
generating, by the system, one or more event representations based on the candidate event components using graph embeddings.

16 . The method of claim 15 , further comprising:
employing, by the system, the one or more event representations to cluster the candidate event components into different event types.

17 . The method of claim 16 , wherein the candidate event components comprise candidate event trigger terms and candidate event arguments respectively associated with the candidate event trigger terms, and wherein the method further comprises:
labeling, by the system, the candidate event arguments with one or more role attributes representative of one or more roles the candidate event arguments play with respect to the different event types, and wherein the generating comprises generating the structured event information based on the different event types, the candidate event trigger terms respectively associated with the different event types, the candidate event arguments respectively associated with the candidate event trigger terms, and the one or more role attributes.

18 . The method of claim 13 , wherein the parsing comprises identifying the candidate event components using abstract meaning representation parsing.

19 . A computer program product for unsupervised event extraction, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processing component to cause the processing component to:
parse unstructured text comprising event information to identify candidate event components; and generate structured event information defining events represented in the unstructured text based on the candidate event components.

20 . The computer program product of claim 19 , wherein the candidate event components comprise candidate event trigger terms and candidate event arguments respectively associated with the candidate event trigger terms, and wherein the program instructions executable further cause the processing component to:
generate event representations based on the candidate event trigger terms and the candidate event arguments using graph embeddings; employ the event representations to cluster the event representations into different event types; and label the candidate event arguments with one or more role attributes representative of one or more roles the candidate event arguments play with respect to the different event types; and generate the structured event information based on the different event types, the candidate event trigger terms respectively associated with the different event types, the candidate event arguments respectively associated with the candidate event trigger terms, and the one or more role attributes.

Assignees

• IBM

Inventors

• Haldar Rajarshi
• Deng Yu
• Wu Lingfei
• Mahindru Ruchi
• Tao Shu

Classifications

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/088

  Non-supervised learning, e.g. competitive learning · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F16/243

  Natural language query formulation · CPC title