Chat text feature classification method and device and storage medium

Abstract

The invention discloses a chat text feature classification method and device, and a storage medium, and the method comprises the steps: preprocessing an obtained a chat text to acquire a word vector;inputting the word vector into a convolution network layer; calculating and generating local feature vectors of the chat text; and connecting the local feature vectors to form a context semantic feature vector; inputting the context semantic feature vector into a deep convolutional neural network to output a first fixed length vector; combining the word vector with a position vector representing the position of each word in the chat text to form a joint word vector; obtaining a second fixed length vector by the joint word vector through a threshold linear unit network GLU and in combination with a multi-core Depthwise convolutional network layer; connecting the first fixed length vector with the second fixed length vector to obtain a multi-level text semantic vector; and inputting the multi-level text semantic vector into a full connection network layer to calculate an output vector; and for the output vector, calculating a classification probability value of the chat text by using a softmax function to obtain a feature category to which the chat text belongs.

Description

technical field [0001] The present invention relates to the field of natural language processing, in particular to a chat text feature classification method, device and storage medium. Background technique [0002] The instant messaging tool represented by WeChat has become an important medium for social interaction, self-expression and sharing of opinions for the vast number of Internet users. As of December 2019, the number of active WeChat users exceeded 960 million, and the number of QQ users exceeded 740 million. Hundreds of millions of active users come from different social and cultural backgrounds, and use social instant messaging tools to communicate in life and work, express opinions, express feelings, and socialize all the time. But at the same time, some illegal and criminal activities are also hidden in it. A large amount of false information, advertising information, Internet fraud and other illegal information are released through QQ and WeChat, seriously end...

AI Technical Summary

Problems solved by technology

[0005] In view of the performance degradation, dimensionality disaster, data sparseness, artificial features that are difficult to describe the hidden semantics of the existing text classification methods on social chat short texts mentioned above, and the use of recurrent neural network methods for text classification and recognition, the calculation Reasoning is slow and cannot be parallelized, etc.

Images