Enterprise public opinion analysis method and device based on self-attention

Abstract

The invention provides an enterprise public opinion analysis method and device based on self-attention. The method comprises the steps that text data are extracted from an obtained Chinese enterprisepublic opinion data set, emotion tags are checked, the text data are preprocessed, stop words, low-frequency words and missing values are removed, the processed text data are obtained, and the text data comprise a title, an abstract and a text of the public opinion data; lexical learning is performed on the processed text data by using a bidirectional GRU algorithm, performing training to obtain alanguage model, and learning is performed to obtain a corresponding word vector; the word vectors are input into a self-attention encoder to learn the relationship between words, and a syntactic-level text representation mode is learned by calculating a self-attention coefficient; and the probability of each sentiment category is calculated through an sentiment analysis classifier according to the syntactic-level text representation mode to obtain an analysis result of enterprise public opinions.

Description

technical field [0001] The present invention relates to the technical field of emotion analysis, in particular to a method and device for analyzing enterprise public opinion based on self-attention. Background technique [0002] The popularization of the Internet has greatly promoted the development of social networks, and the management of public opinion has been paid more and more attention by the relevant departments of enterprises. In the face of turbulent public opinion, departments that have not dealt with similar incidents often seem at a loss. Here are the basic principles that need to be followed in the handling of public opinion. Reasonable monitoring of corporate public opinion data can not only maintain a good reputation of the company, but also help the company adjust its products or strategies. How to quickly mine emotional tendencies from network texts can provide effective information for government, business and individual decision-making. Help has become a...

AI Technical Summary

Problems solved by technology

The dictionary-based method specifies a series of sentiment dictionaries and rules, performs paragraph parsing, syntactic analysis, and calculates the sentiment value of the text, and finally uses the sentiment value as the basis for the text's sentiment tendency. Formulation, the rules in different fields are not the same, the workload is very huge, and the variability of language makes some rules not applicable

There are many methods based on scoring. Bayesian, support vector machine and neural network are used to classify texts, and finally the scoring system is used to output public opinion sentiment labels. A threshold range needs to be set in advance. If the result output by the scoring system is greater than this range, If it is less than this range, it will be judged as negative emotion. If it is within this threshold range, it will be judged as neutral emotion. The biggest disadvantage of this method is that it is not flexible, and the algorithm has certain defects. The characteristics of emotional data may vary greatly. Converting different degrees of characteristics into the specified threshold range will cause a certain degree of deviation, resulting in inaccurate results of the final public opinion analysis.

Images