In this paper, we delve into opinion mining and sentiment analysis of customer reviews posted on online e-Commerce portals such as Amazon.com. Specifically, we look at novel ways of automatic labelling of data for customer reviews by looking at the number of helpful votes and subsequently determine hidden factors that can explain why a customer review is more helpful or trustworthy in contrast to others. We further utilize the factors identified by Multiple Factor Analysis to training Logistic Regression and Support Vector Machine (SVM) models for classifying reviews into trustworthy and non-trustworthy. Experiments show the effectiveness of our proposed approach.
展开▼
