An Efficient Model for Text Sentiment Analysis
Yong Huang, Siwei Liu · 2019
With the rapid development of text sentiment analysis, the demand for automatic classification of electronic documents has grown by leaps. In this paper, we propose a technique for text sentiment classification using term expression attention along with sentiment word embedding vector, which can represent both semantics and sentiment information. Firstly, the sentiment vector is constructed by Long Short-Term Memory(LSTM) and attention mechanism. Then, it is combined with the semantics vector obtained by Word2Vec to form sentiment word vector. Meanwhile, we introduce the concept of expression attention, which is inspired by term frequency- inverse document frequency (TF-IDF). Finally, the sentiment of the text is classified by TextCNN, whose input is sentiment word vector. The model is evaluated by using small, widely used sentiment and subjectivity corpora. The results show that the proposed model out-performs several methods without sentiment vectors or using sentiment dictionaries to generate sentiment vectors.