Sentiment Analysis of Chinese Short Text Based on Multiple Features

With the rapid development of mobile internet and the increase of social platforms, higher performance requirements are put forward for sentiment analysis of Chinese short texts. The traditional deep learning models based on CNN, LSTM and other frameworks face the problem of unable to extract all the effective information contained in the text because of the single direction of text parsing. However, complex model which combines framework in serial way exist a problem that cannot be ignored, that is it can not get effective training. To solve the above problems, this paper proposes an interpretable emotion analysis framework MIX-CNN-BiLSTM-Attention-Transformer (MCBAT Model), which can extract different features from multiple models. From the three dimensions of the fixed collocation of words, the context information and the importance of words in the text, CNN, BiLSTM-Attention and Transformer model are used to extract the above three different features. After vector splicing, the classification results are obtained by the classifier through the full connection layer. The accuracy and stability of the MCBAT model are improved compared with other classical emotion analysis models (CNN, BiLSTM, CNN-BiLSTM, etc.) and LSTM-LDA model. The model based on multi feature consideration is of great significance to emotion analysis task, and provides a method support for further development in the future.