Adaptive Model for Sentiment Analysis of Social Media Data Using Deep Learning

Due to inception of Web 2.0 and increased dependency and freedom to share views, thoughts, opinions on social media, there is high rise in generation of digitized, opinionated social media data. Online forums, blogs, micro-blogging sites, shopping sites, etc. are inundated with mammoth data. This data from multiple domains needs to be extracted and analyzed in order to get the notion of timely insights, and ongoing trends. Many sectors like industries, academia, government and firms are interested to know the sentiments of people towards launched schemes, sales, products, service, policies, rules, etc. to make decisions. Therefore, inferring ongoing trend of topics and finding associated sentiments from huge scale of social data in an automated manner is the need of the hour. This paper proposes an adaptive model for aspect based sentiment analysis of social media data with deep learning approach. Unlike existing methods, our approach performs the task of aspect modeling and sentiment analysis simultaneously using latent semantic indexing with regularization and long short term memory model respectively. The proposed model does not require feature engineering and it is adaptable to datasets of varied domains.