Bidirectional-GRU Based on Attention Mechanism for Aspect-level Sentiment Analysis

Abstract

Aspect-level sentiment analysis is a fine-grained natural language processing task. For traditional deep learning models, they cannot accurately construct aspect-level sentiment features. Such as, for sentence of the movie is very funny, but seats in theater is uncomfortable. For movie, polarity is positive, but it is negative for seats. To deal with this problem, we propose a bidirectional gated recurrent units neural network model that integrates attention mechanism to solve task of aspect-level sentiment analysis. The attention mechanism can focus on different parts of a sentence when sentence has several different aspects. Because we use a bidirectional gated recurrent unit, we can get independent context semantic information and get deeper aspect sentiment information from front and back, so that we can deal with specific aspect sentiment polarity. Finally, we experiment on SemEval-2014 dataset and twitter dataset, result of experiments verified effectiveness of attention-based bidirectional gated recurrent unit on aspect sentiment analysis. The model achieves good performance at different datasets and has further improvement comparing to previous models.