Sentiment Analysis Using A Vector Clustering Approach

Analyzing sentiments plays an attractive role on text mining, information retrieval, and natural language processing tasks. In this thesis, we start with a word vector representation and present a cluster-based method to classify Amazon reviews. Our proposed method is able to remove irrelevant parts within a review and capture the rich sentiment cluster. We experiment with several topic domains of Amazon reviews and use Support Vector Machines on our feature vectors. We also compare our method with lexicon approaches and a Na????ve Bayes classifier. Our method achieves a 65% overall accuracy in a two classes experiment and a 30% overall accuracy in a five classes experiment. This thesis contributes to the sentiment analysis research. Specifically, it introduces an approach to combine the word vector representation with clustering algorithms. Experimental results also indicate that the proposed method can improve the performance of classification in a domain-independent manner