Social media sentiment analysis and topic detection for Singapore English

Social media has become an increasingly important part of our daily lives in the last few years. With the convenience built into smart devices, many new ways of communicating have been made possible via social-media applications. Sentiment analysis and topic detection are two growing areas in Natural Language Processing, and there are increasing trends of using them in social media analytics. In this thesis, we analyze various standard methods used in supervised sentiment analysis and supervised topic detection on social media for Colloquial Singapore English. For supervised topic detection, we created a naïve Bayes classifier that performed classification on 5000 annotated Facebook posts. We compared the result of our classifier against open source classifiers such as Support Vector Machine (SVM), Maximum Entropy and Labeled Latent Dirichlet Allocation (LDA). For supervised sentiment analysis, we developed a phrasal classifier that analyzed the polarity of 425 argumentative Facebook posts. Our naïve Bayes classifier gave the best accuracy result of 89% for supervised topic detection on two-class classification and 57% accuracy for our six-class classification. For our supervised sentiment analysis, our phrasal sentiment analysis classifier obtained an accuracy of 35.5% with negative polarity class achieving a high precision of 94.3%. ; Approved for public release; distribution is unlimited. ; Civilian, ST Engineering, Singapore ; http://archive.org/details/socialmedisentim1094538272