Sentiment Classification Based on Clause Polarity and Fusion via Convolutional Neural Network
Bin Jiang, Hefeng Zhang, Chao Lv, Chao Yang · 2018
This paper presents a novel sentiment classification method based on the clause polarity and fusion via convolutional neural network. Firstly, the sentiment polarity of multiple clauses that make up the original sentence are calculated by the neural network, then the outputted polarity of the clauses are merged by polarity fusion rules to calculate the sentiment polarity of the original sentence. Unlike traditional convolutional neural network which focuses on the most critical sentiment features of a sentence, the model proposed in this paper could consider the effect of sentiment features in each clause on the classification results. It will make the model more effective on the complex classification scene such as the fine-grained sentiment analysis. The experiments results on the two public datasets as SST-1 (5-class) and SST-2 (2-class) datasets show that the model proposed in this paper is feasible and effective.