Scalable Sentiment Analysis for Microblogs Based on Semantic Scoring

Scalability is one of the main challenges of social media analyses such as sentiment analysis. Micro logs as emerging opinion sharing platforms require new approaches that are more scalable and accurate. In this paper, we propose, implement, and evaluate SSSA, a Semantic Scoring Sentiment Analysis service, which matches the demand of scalability and efficiency of a sentiment analysis system for a large volume of micro logs. We first build an automated tweet tagging system to alleviate the cost of building a dataset for training. Our approach then calculates semantic scores of words in tweets. These scores are the baseline for the proposed subjectivity classification and feature selection steps. Finally, we build a polarity classifier to estimate the polarity of subjective tweets on top of them. Experiments show that SSSA, compared to conventional approaches that based on emoticon, hash tag, or semantic scoring, has improved the scalability and accuracy of micro log sentiment analysis.