S-Sense: A Sentiment Analysis Framework for Social Media Sensing

Due to the explosive growth of social media usage in Thailand, many businesses and organizations including market research agencies are seeking for tools which could perform real-time sentiment analysis on the large contents. In this paper, we propose S-Sense, a framework for analyzing sentiment on Thai social media. The proposed framework consists of analysis modules and language resources. Two main analysis modules, intention and sentiment, are based on classification algorithm to automatically assign appropriate intention and sentiment class labels for a given text. To train classification models, language resources, i.e., corpus and lexicon, are needed. Corpus consists of a collection of texts manually labeled with appropriate intention and sentiment classes. Lexicon consists of both general terms from dictionary and clue terms which help identifying the intention and sentiment. To evaluate performance and robustness of the analysis modules, we prepare a data set from Twitter posts and Pantip web board in mobile service domain. The experiments are set up to compare the performance between two different lexicon sets, i.e., general and clue terms. The results show that incorporating clue terms into feature vectors for constructing the classification models yield significant improvement in terms of accuracy.