Short text sentiment fine classification method based on hybrid classifier

Abstract

The invention discloses a short text sentiment fine classification method based on a hybrid classifier. A support vector machine classifier, a long and short memory network classifier and a dictionary-based classification method are combined to form a hybrid classifier. Each classifier is trained by using a small number of training samples; then the emotion categories of unmarked samples are predicted by using the classifiers, the confidence coefficients of texts are calculated by using a weighting formula, the texts with higher confidence coefficients are added into a training set, the classifiers are trained by using a new training set, and iterating is performed to form the classifier with the best effect for classification. The method has the advantages of being good in classificationeffect and small in marking training set requirement.

Description

technical field [0001] The invention relates to the field of natural language processing, in particular to a short text sentiment subdivision method based on a hybrid classifier. Background technique [0002] With the development of Internet technology, more and more information platforms have emerged, covering all aspects of daily life, such as social information, hot news, shopping, entertainment, etc. People publish their own information on various platforms according to their hobbies. However, there are also great differences in the attitudes and opinions expressed by each person towards various phenomena in life. This difference contains great commercial and social value. Therefore, sentiment analysis of this kind of text has very important significance and application value. [0003] In recent years, as the academic community has paid more and more attention to text sentiment analysis, research institutions at home and abroad have continuously invested manpower and ma...

AI Technical Summary

Problems solved by technology

[0005] In practical applications, although semi-supervised learning can reduce the demand for the number of labeled training sets, due to the small number of labeled text training sets, it is impossible to train a single classifier with a high accuracy rate, which leads to the classification of classifiers. The classification accuracy of unlabeled samples is low, and the samples are marked with wrong emotional labels. These wrong labels will be added to the training set, which will cause noise pollution to the training set and affect the training results of the classifier.

Images