Attention-Based Bi-DLSTM for Sentiment Analysis of Beijing Opera Lyrics

Abstract

We propose a sentiment analysis model based on Bi-DLSTM to solve the problem of sentiment analysis of Beijing Opera lyrics. A Bi-LSTM network with dilated recurrent skip connections (Bi-DLSTM) is introduced in this model, which can improve the ability to exact long-sequence information. The proposed model can learn the dependence of long sequences in different time dimensions and effectively improve the semantic extraction performance of lyrics. The attention mechanism is introduced to ensure the recognition of the more important words in the text sequence, which further improves the performance of the network. In order to solve the problem of lack of data on lyric sentiment analysis on the Internet, we build a dataset that can be used for lyric sentiment analysis. This paper completes multiple experiments on four datasets and verifies the effectiveness of the proposed model.