R-SA: A Rule-based Expert System for Sentiment Analysis

Abstract

Sentiment analysis plays a very important role in finding out opinion of the user. Due to the complexity and variety of language grammars and heterogeneous data, sentiment analysis from text is still considered to be the most complex problem till date. For the past few years, classical approaches have been overlooked over newer machine learning techniques which are being used across a myriad of domains. These techniques are typically classified into supervised and unsupervised categories. There are many approaches under supervised categories still under development and completely depending on an unsupervised approach is detrimental to accuracy. Therefore, classical supervised rule-based approaches cannot be ignored. The study focuses on text-based sentiment analysis using supervised classical approach. A novel expert system for sentiment analysis developed in Java is proposed which is a unified framework for rule base and rule engine that is reliable, portable, interoperable and easy to modify for any generic use. The rules are designed based on standards and to enhance the knowledge base, cross-domain inputs are used. The system is compared with existing rule-based engines and results prove that R-SA outperforms existing state-of-the-art classical approaches like Jess, IBM MS-BR and Drools thereby providing a robust and high-performance expert system for textual sentiment analysis and many other applications.