Sentiment of Primary Features in Aspect Based Sentiment Analysis of Hindi Reviews

Aspect Based Sentiment Analysis (ABSA) is a kind of Sentiment classification which allows us to associate sentiments to primary features (Aspect Category) of the object or entity. First, we derive Sentence Vector and sentiment features for a review sentence. Then, we propose algorithms for extracting n-gram features (NGram) and Category Association Word features (CAW). This paper experiments with these four sets of features for building the Ensemble Model for Category-based polarity determination. The result compared with the state-of-the-art reveals that there is improvement in performance for major aspect categories. We also derive an extended Hindi ABSA Category dataset and compare our models’ results. The results show that the accuracy of this subtask ranges from 50% to 76% among four significant domains.