Public opinion sentiment classification method based on deep learning

Abstract

The invention discloses a public opinion sentiment classification method based on deep learning, and applies an XLNet + BiLSTM + Attention model to perform sentiment classification. The method mainly comprises the following steps: data preprocessing: preprocessing microblog hot event comment data; data pre-training: putting the data into an XLNet model, and extracting XLNet word vector representation; feature extraction: inputting the word vectors into a BiLSTM model to obtain text context information features; attention operation: operating an Attention mechanism to extract a deeper feature vector through a weight value; and sentiment classification: performing normalization by using a softmax function, and predicting the sentiment tendency of the speech. The model provided by the invention mainly aims at the problems that data are different and fine tuning is inaccurate during BERT model training and testing, the existing emotion classification is mainly divided into two classes: positive and negative classes, emotion judgment is relatively rough, and training and testing data in an XLNet + BiLSTM + Attention model are not different. According to the result, various emotions rich in the text can be analyzed, and the accuracy of emotion prediction is effectively improved.

Description

technical field [0001] The present invention relates to the field of emotion classification of natural language processing and the field of deep learning, in particular to a deep learning-based public opinion emotion classification method. Background technique [0002] For the problem of sentiment classification in the field of natural language processing, people's research methods range from public opinion sentiment analysis based on sentiment lexicon to machine learning-based public opinion sentiment analysis. Now the most popular is the public opinion sentiment analysis method based on deep learning. The deep learning method saves a lot of labor time and does not need to perform feature extraction operations one by one. Neural networks can actively extract feature vectors through automatic learning. Neural network algorithms such as CNN, RNN, and LSTM are commonly used algorithms in sentiment classification problems. However, the accuracy of these neural network method...

AI Technical Summary

Problems solved by technology

However, the RNN algorithm will have the problem of gradient disappearance or explosion during backpropagation.

Images