Deep sequential model for review rating prediction

Abstract

Sentiment Analysis of review data is becoming an important task to understand the needs and expectations of customers. The challenges that lie in review sentiment analysis is capturing the long term dependencies and intricacies to model the interrelationship between the sentences of the review. In this work, we address the problem of review sentiment analysis using deep sequential model viz. Long short term memory (LSTM) and Gated Recurrent Neural Network (GRNN). LSTM, a variant of RNN is used to process the sentences to a fixed length vector. GRNN is used to capture the interdependencies that exist between the sentences of a review. The combination of LSTM and GRNN shows good performance on Amazon Electronics dataset.