Electronic official document entity extraction method

Abstract

The invention provides an electronic official document entity extraction method. The electronic official document entity extraction method comprises the following steps: A, preprocessing; B, constructing features; C, training an entity extraction model; D, obtaining a corpus; E, obtaining a word vector; F, training an algorithm model. According to the method, a traditional sequence labeling algorithm and a deep learning algorithm are combined, the advantage that a traditional sequence algorithm needs less corpus labeling is utilized, a semi-supervised method is adopted to expand corpuses, andthe problem that time and labor are wasted when a large number of corpuses need to be manually labeled in the deep learning algorithm is solved. Maximum forward and reverse dictionaries, syntax and semantic features are added into the CRF model, and front and rear boundary word features of entity words are fully considered, so that the algorithm has generalization ability. A dilated CNN and BiLSTM-CRF are combined, the dilated CNN takes a character-level vector and a character-level position feature as external features, and the external features and a part-of-speech vector are spliced into aword vector, so that more semantics and up-and-down related information can be expressed to a certain extent.

Description

technical field [0001] The invention relates to an electronic document entity extraction method, in particular to an electronic document entity extraction method based on a semi-supervised dilatedCNN-BiLSTM-CRF. Background technique [0002] Official documents are written by specific departments and agencies, which not only have legal effect as evidence of policy activities, but also contain a large amount of professional policy knowledge, which contains a large amount of entity information, such as: name of person, place name, issuing agency, organization name, name of scientific research project, time , events, etc., as long as they are special text fragments required by the business, they can be called entities. Extracting these entities or concepts is the premise of official document word segmentation, dependency syntax analysis, and grammatical analysis. work plays an important role. Therefore, how to more accurately identify entities from official documents has become...

AI Technical Summary

Problems solved by technology

However, deep learning methods generally require a large amount of labeled corpus, otherwise it is very easy to overfit and fail to achieve the expected generalization ability

Images