BiLSTM Model With Attention Mechanism for Sentiment Classification on Chinese Mixed Text Comments

Abstract

Sentiment classification has become a significant research topic in natural language processing. As the most popular research method of sentiment classification, deep learning has been applied to various experimental datasets by numerous scholars. However, sentiment classification suffers from a number of drawbacks including poor effect and insufficient accuracy in current vertical field. Moreover, it has not been applied on Chinese mixed texts, including both long and short texts. Therefore, the present study aims at proposing an improved BiLSTM-Attention model through which features can be extracted more effectively. The proposed model can be argued to resolve the problem of insufficient dependence on long texts by applying the Bi-directional Long Short-Term Memory (BiLSTM) model. Furthermore, important textual information can be obtained by the attention mechanism. To this aim, online shopping comment datasets were investigated and a multiclassification evaluation index was applied to evaluate the model. The findings of the study corroborated the effectiveness of the proposed approach. To be more exact, the accuracy of the sentiment classification of mixed texts and long texts were 0.9280 and 0.9358, respectively. In addition, in terms of classification performance, the practical effect of the proposed method was observed to be more advantageous than other sentiment classification methods. Finally, enjoying a good domain extensibility, the study was found to make an important contribution to business development, as well.

Keywords

• Sentiment classification
• Chinese mixed test
• BiLSTM
• attention mechanism