Document Level Polarity Classification With Attention Gated Recurrent Unit

Reviews can be categorized into two extreme polarities, that is, positive or negative. These reviews from different consumers on a product or service can help a new consumer to make a good decision. Document level sentiment classification aims to understand user generated content or opinion towards certain products or services. In this paper, we propose a recurrent neural network model in classifying positive and negative reviews using gated recurrent unit and attention mechanism. Effectiveness of our proposed model is evaluated using Yelp Review dataset obtained from Yelp Dataset Challenge. Experimental results show that our proposed model can outperform existing models for document level sentiment classification.