Sentiment analysis of movie reviews based on pretraining and dual branch coding

Abstract

Due to the high dimensionality and sparsity of text, and to the complex semantics of the natural language, sentiment analysis of movie reviews presents difficult challenges. In o-rder to solve these problems, a novel architecture which conta-ins a robustly optimized bidirectional encoder representations from transformers pretraining approach (RoBERTa), a bidir-ectional long short-term memory (BiLSTM), a temporal conv-olutional network (fCN), a convolutional layer and an attenti-on mechanism is proposed in this paper. The proposed archit-ecture is called dual branch feature coding network based on RoBERTa (DBN-Ro). In DBN-Ro, the RoBERTa pretraining model is used to obtain the word embedding and the dual bra-nch network is used to extract the contextual semantic and multi-level representations. The convolutional layer is used to perform dimensionality compression on the stitched vectors. The attention mechanism is introduced to highlight important features. The DBN-Ro is experimentally validated on two mov-ie reviews, the results shows that the it can achieve better pre-diction results in sentiment analysis of movie reviews tasks.