Sentiment classification method and system

Abstract

The invention is suitable for the technical field of the computer, and provides a sentiment classification method and system. The method comprises the following steps: according to a rhetorical structure theory, analyzing a text to be classified to obtain a rhetorical structure parse tree; obtaining the initial vector of each node in the rhetorical structure parse tree, wherein the node comprises an input gate, an output gate, a memory cell, a hiding state and a forgetting gate; according to the hyperbolic curve tangent function value of the output gate and the memory cell of the node, carrying out dot product to obtain the hiding state of the node; and according to the hiding state of the node, carrying out sentiment classification on a classifier function. The text to be classified is constructed into the rhetorical structure parse tree, each layer in the rhetorical structure parse tree is provided with two node fragments, each node is provided with an own forgetting gate, child node information is selected through the forgetting gate in a learning process, a cell state is continuously updated, unimportant information is abandoned, core contents are added, and semantic feature expression is improved so as to improve classification accuracy.

Description

technical field [0001] The invention belongs to the technical field of computers, and in particular relates to an emotion classification method and system. Background technique [0002] Due to the problem of gradient disappearance in Recurrent Neural Networks (RNN, Recurrent Neural Networks), in recent years, a chain-structured long-short-term memory network (LSTM, Long short-term memory) has been proposed and widely studied. It adds a memory cell structure to the original RNN to store information. This improvement makes LSTM have a strong ability to protect sequence information over time, so it can capture long-term, long-distance dependencies. Solved the gradient vanishing problem of RNN. However, the previous research on LSTM was based on the linear chain structure of time series. Later, after research, the chained LSTM was extended to the tree structure and constructed through the syntax analysis tree structure, which improved the expression of semantic features. These ...

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to provide a method and system for emotion classification, aiming to solve the problem that the semantic features learned in the prior art are not sufficient and accurate, resulting in poor effect of emotion classification

Images