Author(s)

Wenjin Chen, Jianhua Wen

Corresponding Author

Wenjin Chen

ABSTRACT

Online ratings and reviews reflect consumer's real evaluation of the purchase, and contain a huge commercial value. After a preliminary understanding of the provided data, we perform sentiment analysis on the review data, fuse and replace words with consistent expressions. We use Poisson regression to explore the relationship between reviews and star ratings, and the results show that they have a positive correlation. At the same time, we utilize the Covariance Weighting method to integrate reviews and star ratings, to build a comprehensive index (RD) that quantifies the direction of consumer feedback. Considering that not every review has the same degree of confidence, we incorporate the following data into our model. Then, we combine these factors into a weighted index (RR) that reflects the reliability of the review, with the help of the propensity score weighting method. Besides, we put consumer feedback data into the model and compare their credibility. Finally, we fuse the above two models, and use the index model to make RD and RR finally generate an index (FR), which clearly shows the reputation of the product after sale. And, put time and FR into exponential smoothing model to get the trend of FR in the given data. To sum up, after making sensitivity analysis, it shows that our model is robust. Besides, our model is feasible and reasonable for solving Sunshine’s problem. Nevertheless, there are also some existing problems like any model. Yet, it will be more powerful after further extension.

KEYWORDS

Sentiment analysis, poisson regression, propensity score weighting method