Publications

Abstract : Online reviews and their star ratings are an important asset for users in deciding whether to buy a product, watch a movie, or where to plan for a tour. The review text is quite explanatory and expressive where the user shares his experiences in a detailed manner. But the star rating attached to a review is often subjective and at times biased to the reviewer’s personality. The priorities given to different aspects or features by a user might in turn influence his rating. This discrepancy in the review content and the star rating makes a significant impact when sentiment is extracted from large voluminous data. This emphasizes the need to build a model to predict the rating based on the textual content in the reviews rather than bluntly following the star rating attached to the review. A novel weighted textual feature method is applied to assign an appropriate score to each review. Sequential Minimal Optimization (SMO) regression is applied to the feature set to predict the review rating which is based purely on the textual content of the review. The predicted rating has been compared with the actual user rating, and an analysis has been carried out to draw insights into the discrepancy. The model has been experimented in the tourism domain, and the results are quite promising. The regression residual value obtained for hotel and destination data is 0.8772 and 0.9503.