A chapter-level event extraction method in the financial field based on article entity word dependencies

Abstract

The invention discloses a chapter-level event extraction method in the financial field based on the dependency relationship of entity words in articles, and designs a structured dependency self-attention mechanism module, which uses the dependency relationship of entity words in articles as a deep learning model for event extraction A kind of input data, which is combined with word-level and sentence-level language features to improve the prediction accuracy of event trigger words and event arguments when the deep learning model extracts financial events. In addition, in the task of extracting Chinese financial events, the present invention proposes 8 different types of entity relationships for the first time, which are used to uniformly represent entity dependencies in articles. The present invention simultaneously builds a set of hierarchical event relationships in the financial field, which are used for models to distinguish similar event types. From system input to system output, the present invention has a clear logical structure, clear layers, detailed system implementation details, and realizes an end-to-end system closed working mode, which is very easy to implement and to be applied on a large scale.

Description

technical field [0001] The invention belongs to the intersecting field of artificial intelligence and finance, and in particular relates to a chapter-level event extraction method in the financial field based on the dependency relationship of article entity words. Background technique [0002] At this stage, artificial intelligence is gradually getting involved in various industries and fields, and provides new possibilities for its development. The financial field, as one of the industries that generate the largest real-time data, has also entered the era of "AI+finance". As one of the important tasks in the financial field, it is very meaningful and valuable to extract major financial events with value from major financial announcements / news. The currently known chapter-level financial event extraction methods are event extraction systems based on traditional rules and traditional machine learning paradigms, which have the following deficiencies: [0003] 1) Most of the ...

AI Technical Summary

Problems solved by technology

[0003] 1) Most of the current financial event extraction methods are mostly based on single-sentence event extraction, which cannot handle chapter-level, multi-sentence financial event extraction;

[0004] 2) The existing chapter-level financial event extraction methods do not consider the relationship information between entities; for example, the meaning of the same entity appearing in different sentences, the semantic information represented by different entities appearing in the same sentence at the same time, etc.

Images