A framework for mining on Twitter data

Abstract

Motivated by the increasing need of information retrieval from social media, a lexicon-based approach Tweet Sentiment Classifier (TSC) is presented to determine sentiment from tweet along with a systematic software for twitter data statistics analysis and topic extraction. The TSC uses annotated dictionaries of words (SentiWordNet) and has a negation detector. While the LDA topic model uses Gibbs Sampling. The entire system is unsupervised. Without the need of training, it has significant advantage on speed comparing to supervised methods. It is robust to provide consistent satisfying results from different topics of twitter data. The performance of the TSC also outperforms one of the baseline sentiment analysis methods.