Chinese comment sentiment classification method and device based on evidence inference rules

Abstract

The invention provides a Chinese comment sentiment classification method and device based on evidence inference rules. The method includes: preprocessing the acquired Chinese comment text; extracting features of the preprocessed Chinese comment text to obtain a vectorized representation of the Chinese comment text; dividing the Chinese comment text into training data and test data, and Use the Chinese review text to train multiple preset base classifiers; use the trained multiple base classifiers to predict the sentiment tendency of the test data to obtain the weight of each base classifier; based on the The weight and reliability of each base classifier are combined with the output probability of each base classifier by using evidence inference rules to obtain the sentiment classification result of Chinese comment text. In this way, the present invention uses evidence inference rules to fuse the output results of different base classifiers, and takes into account the weight values ​​and reliability of different classifiers, so that the final fusion result is more reliable and accurate.

Description

technical field [0001] The invention relates to the technical field of data mining, in particular to a method and device for classifying Chinese comment sentiment based on evidence reasoning rules. Background technique [0002] The problem of judging the sentiment polarity of Chinese text is a typical binary classification problem, involving many fields, such as natural language processing, artificial intelligence, text mining, etc. For the review text corpus of a specific industry, the emotional polarity judgment is mainly to use computer technology to detect and extract the user's subjective information, opinions and emotions in the text document, so as to mine the user's emotional tendency. [0003] At present, the existing popular sentiment analysis method is conceived as follows: convert the text sentiment analysis problem into a classification problem, and through the steps of text preprocessing, sentiment feature extraction and feature weight calculation, the text is ...

AI Technical Summary

Problems solved by technology

[0005] However, when using the traditional TFIDF method to calculate the weight of Chinese text emotional features, the distribution of feature words among different categories is not considered, and it is easy to assign high weights to those feature words that are evenly distributed among categories and are not strong in class distinction.

Therefore, inaccurate feature weight values ​​will affect the final text sentiment classification effect

And, when using an ensemble algorithm to solve the problem of Chinese text sentiment classification, if the voting method, Boosting and other algorithms are used to integrate multiple classifiers for text sentiment classification, although the instability of a single classifier can be well improved, these methods are only Simply considering the support degree of the classifier to the category of the sample, ignoring the probability that the sample does not belong to the category in the output result of the classifier, and not making full use of the output information of the classifier, these integrated classification methods sometimes do not Significantly improved classification results

[0006] In addition, if the evidence theory is used to integrate multiple classifiers for text sentiment classification, although this method takes into account the uncertainty of the output probability of the classifiers, it does not take into account the weight and reliability of each classifier, and it is easy to integrate high conflicts. When the classifier outputs results, there are uncommon sense results, resulting in poor classification results

Images