Abstract: To address the problems that the existing methods cannot comprehensively consider the semantic information of text in terms of syntactic structure, contextual information and local semantic features, a feature fusion-based Chinese text sentiment analysis method is proposed. Firstly, Jieba word segmentation tool is used for word segmentation and part of speech tagging of the review text, and a word vector training tool GloVe is used to obtain pre-trained word vectors with part of speech. Then, the word vector is used as the input of BiGRU and TextCNN with self attention respectively. The global features are extracted from the syntactic structure and the contextual information of the text using BiGRU with Self-Attention, and the local semantic ones are extracted using TextCNN. Finally, the global features and local semantic features are integrated, and Softmax is used for text sentiment classification. The experiment result shows that the proposed method can effectively improve the accuracy of text sentiment analysis.

Key words: sentiment analysis of Chinese text, feature fusion, feature extraction, semantic feature, self-attention mechanism, deep learning hybrid model