Feasibility of Using Rating to Predict Sentiment for Online Reviews

More consumers depend on online recommendation for products before making purchase decision. Comments or ratings given by other online users are used as recommendation and help consumers to make informed decision. Giving rating is a more convenient way to express sentiment toward a product while comments provide details and allow for subjective judgment. This paper investigates the problem of using rating alone to infer the actual sentiment polarity of the raters towards a product. In particular, the experiment attempts to discover whether the lower ratings (1 or 2 in 5-point scale) are more associated to negative polarity while the higher ratings (4 or 5 in 5-point scale) are more associated to positive polarity as universally assumed. A lexical based sentiment analysis approach is used to determine sentiment polarity of each textual comment. The results showed that higher ratings could indicate positive sentiment but it is not the case for the lower ratings in representing negative sentiment.