SENTIMENT ANALYSIS OF USER REVIEW TEXT THROUGH CNN AND LSTM METHODS

Abstract

Sentiment analysis, a process of natural language processing, is gaining popularity these days due to hike in the number of Internet users. The Internet users put up their opinions in the form of reviews regarding some contexts like products, services, government, politics, medicine, and entertainment world to name a few. The evaluation of correct sentiments out of web text is one of the main focuses of business houses or large organizations. Several machine learning approaches are used to address the issue of sentiment prediction from raw web text. This paper contributes to the existing methods by proposing a combination of convolutional neural networks and long-short-term-memory network for evaluation of appropriate sentiments. Two open source datasets are extracted in the pre-processing phase of proposed model. The major contribution of this research work comes from the pre-processing phase of data where a novel zero-padding method is used for normalization of word features before configuring neural networks. The proposed system outperforms the baseline classifiers almost 13% higher than the best performed baseline system for the first dataset and 3% improvement in accuracy is observed for the second dataset.