Sentiment Orientation from Code-mixed Social Media Data

Abstract

Collecting and evaluating data is becoming an effectively admissible challenge in the highly connected world. In the 21st century, with the advent of social networks getting popular, the social media information is getting archived at alarming rates. The use of local language in informal fashion is very common on social media platform. In natural language processing, Sentiment Analysis (SA) is a specialized process of determining user orientation from opinion data floating on social web. Code-mixed social media data in specific is challenging to process, due to mixing of varied languages used to portray the linguistic efficiency. In this paper, we propose a model called Code-mixed Sentiment Analyzer (cmSentiAnalyzer) to derive sentiment orientation from code-mixed sentences. Our proposed model has used language features across code-mixed languages to map the words occurring in different languages to a common space. Our experiments reveal that cmSentiAnalyzer outperforms baseline approaches in sentiment analysis for code-mixed text by 2% in accuracy and 89% of average precision.