Tibetan Comment Text Sentiment Recognition Algorithm Based on Syllables

Abstract

Text sentiment recognition is to analyze and process the text with sentimental color, and then divide the text into different sentiment categories. This paper uses Tibetan syllables as the basic features of the text. The first is to obtain the contextual global sequence features of Tibetan comment texts through the bidirectional recurrent neural network model. Then through the Self-Attention mechanism, further obtain the relationship features between the Tibetan syllable feature units in the sequence. Finally, convolutional neural network models with different convolution kernels are used to obtain fine grained partial features of the text. After completing the text representation, input it to the fully connected layer for sentiment recognition. This paper improves the performance of text sentiment recognition by obtaining richer and comprehensive semantic features in Tibetan comment texts. On the Tibetan-Chinese bilingual comment corpus, a variety of deep learning models are compared. The model proposed in this paper effectively improves the accuracy of text sentiment recognition.