MODELLING OF CONTEXT-BASED SENTIMENT ANALYSIS AND ITS APPLICATIONS

Abstract

Social media mining has proven valuable in numerous research areas as a pragmatic tool for public opinion extraction and analysis. Sentiment analysis (SA) addresses the dynamics of complex socio-affective applications that permeate intelligence and decision making in the sentient and solution-savvy social web. It encompasses investigation of both opinion and emotion within the content. Having started with simple polarity detection, contemporary SA has advanced to a more nuanced analysis of context, affect and emotion sensing. Existing sentiment analysis techniques quite efficiently capture opinions from text written in syntactically correct and explicit language. However, while dealing with the informal social data, limitations have been observed in performance of sentiment analysis techniques. Understanding the pragmatics, emotion, cognition and behaviour are key to accurate SA. Ongoing research shows that some of the issues pertaining to natural language use can be resolved by adding extra information (i.e. context) to the process of SA. This research primarily aims to find out the types of contextual information which can be extracted from social media and can be applied to improve results of SA. In this direction, a multi-faceted conceptual framework for context has been built. It defines types of context that can be used in SA. The concept of context was then applied for building a model for contextual SA. In addition to this work, the contextual framework work dealing with single modality (textual data) has been extended to deal with multiple modality data. This conceptualization of ‘context’ was further applied for detection of specialized sentiment like sarcasm with improved accuracy. Sarcasm detection has been carried out for both multiple modality data and multilingual data. This work presents a learning model for real-time sarcasm detection in Hinglish (Hindi +English) code-switch dataset. The empirical analysis has been carried on the available benchmark datasets and also on created datasets. The results have been evaluated using standard classification metrics and proposed techniques have been compared extensively with the existing state-of-the-art. The research affirms that the use of appropriate context information can help in improving the accuracy of sentiment classification and that there is a consistent need to comprehend and apply the multi-faceted concept of context in social data for realtime intelligence.