Sentiment word extension-based short text sentiment classifying method

Abstract

The invention relates to a sentiment word extension-based short text sentiment classifying method, and belongs to the technical field of computers and information science. The sentiment word extension-based short text sentiment classifying method comprises the following steps: first, segmenting a comment text into a sentence set, and dividing words and labelling parts of speech by utilizing a jieba word dividing tool to obtain a pre-processed result; then, aiming at each short text comment, acquiring the word vector of each word by using Wikipedia corpus training Glove, calculating the semantic similarity of other words and the primary sentiment features with the parts of speech of N, V, Adj and Adv by utilizing the word vectors, expanding the words with similar semantics to a primary sentiment feature set; next, proposing DF-TF-MI, performing feature dimension reduction by improving a conventional feature dimension reduction method by utilizing interlexical statistical features to obtain a low-dimension feature set, and weighting the sentiment features; finally, performing sentiment tendency classification on the obtained feature vectors through an RADA algorithm formed by weak classifier weighting. According to the sentiment word extension-based short text sentiment classifying method, the problem that unregistered words exit in a sentiment dictionary is solved, meanwhile, the problem of sparse sentiment features caused by few effective sentiment words of the short text comment is solved, and the performance and accuracy of sentiment tendency analysis are improved.

Description

technical field [0001] The invention relates to a short text emotion classification method based on emotion word expansion, and belongs to the technical field of computer and information science. Background technique [0002] Chinese short texts are shorter in length and rich in less information. However, the short text user comment information contains the user's emotional tendency. Through short text sentiment analysis, users' true opinions and attitudes hidden under the surface text can be effectively mined. Therefore, the present invention will provide a short text sentiment classification method based on sentiment word expansion to improve the accuracy of comment text sentiment classification, thereby enhancing the practical value of systems such as user evaluation analysis and product recommendation, which has important theoretical significance and commercial value. [0003] The basic problems that need to be solved in the short text sentiment classification method b...

AI Technical Summary

Problems solved by technology

[0003] The basic problems that need to be solved in the short text sentiment classification method based on the expansion of sentiment words are: 1. Short text comments are short, and there are few sentiment feature words, which may easily lead to sparse sentiment features; 2. The existing short text sentiment analysis methods are not effective in feature dimension reduction Good; 3. The choice of classifier in the supervised sentiment analysis method has a great influence on the accuracy of sentiment classification

However, the selection of classifiers in the supervised sentiment analysis method has a great influence on the accuracy of sentiment classification.

[0008] To sum up, the existing models do not solve the problem of sparse emotional features of short texts. In addition, the existing short text sentiment analysis methods are not effective in feature dimensionality reduction. The choice has a greater impact on sentiment classification accuracy

Images