Hybrid Approach Framework for Sentiment Classification on Microblogging

Microblogging is used widely to express opinions toward an entity, knowing sentiment polarity can get benefit for decision making, planning, and visualization and so on. Outdated training data along with the nature of Microblogging which is short and noisy cause low accuracy. Existing approach requires human effort to manually label huge training data. To tackle these problems, we proposed a framework that used a hybrid approach between lexicon-based approach and machine learning approach. SentiWordnet has been used to automatically label training data and then using Support Vector Machine for sentiment classification. We study two scoring mechanisms for labeling training data, Word Sense Disambiguation and Non Word Sense Disambiguation. The framework also used MapReduce for computing large dataset. The result shows that Non Word Sense Disambiguation is optimal for this framework. The framework is functional, more automatically and less human efforts.