Comment Texts Sentiment Analysis Based on Improved Bi-LSTM and Naive Bayes
Renju Lin · 2022 International Conference on Data Analytics, Computing and Artificial Intelligence (ICDACAI) · 2022
With the development of the Internet, various social media have attracted many people to share their thoughts and describe their feelings. Most of these comments are emotional. In the era of big data, we can discover the hidden value through massive data. Like these comments, we can analyze them through sentiment analysis to obtain the emotions contained in people's comments. The information obtained from these analyses can help enterprises better serve people to achieve corporate development and help social media carry out online community governance. Among the methods of emotion analysis, deep learning has been rapidly developed and widely applied in recent years, such as RNN and LSTM. This paper uses bidirectional LSTM to capture bidirectional semantic dependence better. Traditional machine learning methods also have specific applications in emotion analysis. Naive Bayes, SVM, decision tree model, and other algorithm models with classification ability can be well applied to emotion analysis. Information gain represents the degree of information complexity (uncertainty) reduction under a condition, and the importance of a word in classification can be judged by information gain. In this paper, we use an improved information gain formula. We screened out important classification features through improved information gain and applied these features to feature extraction of the naive Bayes model. Compared with the naive Bayes model, the accuracy of the improved naive Bayes model has been improved to some extent. We also combine the improved information gain model. We use bidirectional LSTM by adding a dimension. Behind the last hidden state calculated by bidirectional LSTM. This dimension is the sum of the forward word improvement information gained in the statement. This Bi-LSTM model, combined with improved information gain, also improves accuracy compared with the traditional Bi-LSTM model.