News event extraction method and device

Abstract

The invention provides a news event extraction method and a news event extraction device. The news event extraction method includes: identifying the category of news to be classified; performing event extraction on news of a preset category according to the category of the news; identifying the subject of the news to be classified; filtering the news that has been subjected to event extraction according to the subject of the news , output the news event finally extracted. The present invention solves the difficult problem of event extraction with some long spans, and also reduces the syntactic analysis requirements of the existing event extraction based on syntactic analysis, and improves the recall rate of event extraction; at the same time, it improves the existing single dependency syntactic analysis with limited accuracy. problem; and, redundant news and redundant events are effectively removed, and the accuracy of event extraction is improved.

Description

technical field [0001] The invention relates to natural language technology, in particular to a news event extraction method based on syntactic analysis and filtering mechanism. Background technique [0002] With the rapid development of the Internet, the information obtained from the Internet is also increasing explosively every day. How to effectively obtain valuable events from this information has become an urgent problem to be solved. Event extraction technology is in such a demand. produced. Event extraction technology can extract events that people are interested in, where, when, and what happened from a large number of texts expressed in natural language. [0003] The main methods of current event extraction and their disadvantages are as follows: [0004] 1. Pattern matching method [0005] The pattern matching method extracts events by defining an event extraction template, which requires a certain amount of professional knowledge to define the template, which i...

AI Technical Summary

Problems solved by technology

However, in the prior art, event extraction based on syntactic analysis often only analyzes sentences directly by using dependency syntax, which requires high accuracy of dependency syntax tools, and generally has a low coverage of event syntax, and it is difficult to achieve Satisfactory accuracy

[0010] On the other hand, the events extracted from the news are generally real, important, and time-sensitive. However, due to the openness of the Internet, the news on the Internet has the characteristics of heterogeneity, redundancy, and dynamic change. News may contain multiple events of interest, but some events do not have the above-mentioned characteristics of authenticity, importance, and effectiveness. How to effectively filter out these redundant events and improve the accuracy of news event extraction is also a problem that needs to be solved question

However, the current methods focus on how to extract events from the news, and do not process these redundant events.

Images