Text classification method based on multi-source features, terminal equipment and storage medium

Abstract

The invention relates to a multi-source feature-based text classification method, terminal equipment and a storage medium. The method comprises the following steps: S1, receiving a text and carrying out word segmentation; s2, obtaining a word attention weight matrix and a word attention weight matrix by adding a self-attention mechanism in the LSTM network; s3, constructing a keyword table, and searching a core keyword table from the keyword table based on a word segmentation result; s4, extracting by adopting N keyword extraction algorithms to obtain N candidate keyword tables; s5, based on the word attention weight matrix and the word attention weight matrix, obtaining an expanded keyword table and an expanded keyword table according to the candidate keyword table; s6, taking all characters and words in the core keyword table, the expanded keyword table and the expanded keyword table as keywords and keywords; s7, performing feature extraction on the keywords and the keywords; and S8, based on the extracted features, predicting the category of the text through a classification network. According to the invention, the text classification accuracy is improved.

Description

technical field [0001] The invention relates to the field of text classification, in particular to a multi-source feature-based text classification method, terminal equipment and storage media. Background technique [0002] In recent years, the mobile Internet has developed rapidly. As of December 2020, the number of Internet users in China has reached 989 million, of which mobile Internet users accounted for 99.7%, and instant messaging apps accounted for 99.2% of mobile Internet users. Communication software represented by QQ and WeChat has become an indispensable part of most people's daily work and life, and it has also become a tool for criminals to publish false news and conduct illegal activities on the Internet. In the field of electronic forensics, keywords are mined from mobile phones, and key topics are quickly and intelligently classified, bringing key progress to the detection of cases. [0003] The diversity of chat topics, the complexity of data, the serious ...

AI Technical Summary

Problems solved by technology

[0005] (1) Traditional method keyword extraction methods include: LDA topic model, TF-IDF, TextRank, etc. and improvements based on these methods. For longer texts, the keyword extraction method based on traditional statistical learning has relatively large noise

The traditional keyword extraction method is single, resulting in limited keyword extraction, especially for the limited ability of keyword extraction in professional fields

[0006] (2) The feature extraction methods based on deep learning include: the method of integrating LSTM and LDA. Although this method can obtain good keyword extraction results based on the word embedding method based on training, each vocabulary pair is classified based on multi-source features. The contribution of is not equivalent, and there is a problem of lexical feature redundancy

Keyword extraction based on the attention mechanism method and the multi-feature fusion method, although to a certain extent makes up for the defect of single etymology for keyword extraction, but the keywords obtained are limited

[0007] (3) Text classification methods based on single etymology or multi-source features based on character features cannot obtain multi-dimensional features at the same time: text classification methods based on lexical features already have certain feature capture and semantic understanding capabilities; The text classification method based on multi-source features based on character embedding can greatly reduce the length of the dictionary, reduce the operation cost, and improve the operation efficiency, but lacks semantic information.

Images