Event extraction method, system and device for unstructured text

Abstract

The invention discloses an event extraction method, system and device for an unstructured text. The method comprises the steps: carrying out the extraction of an event in the unstructured text based on a pre-constructed event intention dictionary library, recognizing the event type of the event, and carrying out the extraction of the event in the unstructured text based on a pre-constructed event role dictionary library, recognizing role elements in the event; According to the event type and the role element, performing structured storage on the event. By configuring the dictionary, the analysis of the event in the question sentence is more flexible, convenient to maintain and easy to expand. The complex procedures of collecting related corpora, labeling, training and the like required by deep learning are avoided.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence natural language processing, in particular to an event extraction method, system and device for unstructured text. Background technique [0002] In the prior art, event extraction is to present unstructured text containing event information in a structured form, and has a wide range of applications in fields such as automatic summarization, automatic question answering, and information retrieval. The core value of event extraction technology is that it can convert semi-structured and unstructured data into a structured description of events, and then support rich downstream applications. [0003] The current entity recognition mainly adopts machine learning and deep learning, but there are shortcomings: [0004] (1) For methods based on deep learning and machine learning, it is first necessary to collect early training corpus and then label them. However, for vertical fiel...

AI Technical Summary

Problems solved by technology

[0004] (1) For methods based on deep learning and machine learning, it is first necessary to collect early training corpus and then label them. However, for vertical fields such as finance, insurance, and petrochemicals, the required labeling corpus is very scarce, which increases the The difficulty of implementing technical solutions; at the same time, the effect of deep learning has a lot of unstable factors, and the accuracy is difficult to control;

[0005] (2) The method based on deep learning and machine learning is cumbersome for the increase of event types and needs to be retrained, which is inefficient

Images