Overlapped event extraction method based on ontology guidance and plan sampling

By fusing ontology knowledge and planned sampling strategies in overlap event extraction, independent sequence encoding is generated, and the problems of slow model training speed and poor generalization ability in the existing technology are solved, and efficient overlap event extraction is achieved.

CN119129587APending Publication Date: 2024-12-13SHENYANG AEROSPACE UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411233359.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-12-13

AI Technical Summary

Technical Problem

In the overlapping event extraction, the existing technology has problems such as slow convergence speed of model training, poor generalization ability, error cascade propagation and label conflict, and ignores the application of event ontology knowledge.

Method used

Using the overlapping event extraction method based on ontology guidance and planned sampling, the semantic coding module, event type discrimination module, trigger word marking module and argument extraction module that fuses ontology knowledge, independent sequence encoding is generated to reduce error levels during training. , and the convergence of the model is accelerated through the planned sampling strategy.

Benefits of technology

It shows excellent overlap event extraction performance in unstructured text, improves the convergence speed and generalization ability of the model, and effectively solves the overlap problem of trigger words and arguments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119129587A_ABST
    Figure CN119129587A_ABST
Patent Text Reader

Abstract

The invention provides an overlapping event extraction method based on ontology guidance and plan sampling, and relates to the technical field of computer natural language processing. A semantic coding module fusing ontology knowledge is adopted, independent sequence codes are generated according to different sub-tasks, and the performance is excellent in an overlapping event extraction task of an unstructured text; in the training process, multi-stage prediction is regarded as sequence output of a joint model, and a plan sampling strategy is used between subtasks, so that cascade propagation of errors in the training process is reduced, the convergence speed of the model is increased, and the training efficiency is improved. And finally, the interaction among the subtasks can be enhanced by combining an event type discrimination loss function, a trigger word marking loss function and an argument extraction loss function. A continuous event type judgment module, a trigger word marking module and an argument extraction module are adopted, and overlapped targets can be extracted on the basis of specific results in the previous stage, so that the problem of overlapping of trigger words and arguments can be effectively solved.
Need to check novelty before this filing date? Find Prior Art