Sentiment Analysis of Customer Reviews Using Robust Hierarchical Bidirectional Recurrent Neural Network

摘要

With tremendous growth of online content, sentiment analysis of customer reviews has become an active research topic for machine learning community. However, due to variety of products being reviewed online traditional methods do not give desirable results. As number of reviews expand, it is essential to develop robust sentiment analysis model capable of extracting product aspects and determine sentiments adhering to various accuracy measures. Here, hierarchical bidirectional recurrent neural network (HBRNN) is developed in order to characterize sentiment specific aspects in review data available at DBS Text Mining Challenge. HBRNN predicts aspect sentiments vector at reviewlevel.HBRNNis optimized by fine tuning different network parameters and compared with methods like long short termmemory (LSTM) and bidirectional LSTM (BLSTM). The methods are evaluated with highly skewed data. All models are evaluated using precision, recall and F1 scores. The results on experimental dataset indicate superiority of HBRNN over other techniques.