Prediction of Sunshine Company's Product Participation in the Online Market based on Consumers' star Ratings and Reviews for Given Products

Abstract

When shopping on line, reviewing product reviews often plays a very important role in making decisions for consumers. We will build models, study the relationship between user reviews and other metrics, and propose appropriate sales strategies for the company.

First of all, we need existing data for processing. We analyzed the star rating, the voting status of reviews, and the sales volume, and divided the reviews into three levels according to the length of the content of the paper. Finally, we reached the following conclusions:

1) The pacifier has the highest average star rating among the three types of products, and the highest sales volume;

2) Three categories of products have the largest percentage of Five Star reviews, and Microwave for two products, one star rating accounted for a relatively large proportion.

Secondly, the TF-IDF algorithm is used to extract keywords from user reviews, and combined with sentiment analysis, an emotional score is assigned to each review. Then, we establish a fuzzy evaluation model and use the reviews that have been scored to analyze the satisfaction of the product. In the end, we got a comprehensive evaluation of the three products, which were all expressed as satisfactory, and the satisfaction degree of the hair dryer was higher than that of the microwave oven and the pacifier.

Using the established model and the obtained data, we can find out the trend of star ratings over time, and filter out the most successful products and the most failed products. In particular, for the most failed products, we can describe keywords based on specific quality descriptions in user reviews, and find the shortcomings of the products to improve them. In addition, we built a model and found that after deleting a specific star rating, other users will be affected when they comment. Finally, we screened out specific keywords of five stars and one star respectively. With reference to their emotional scores, we can find that there is a high positive correlation between specific keywords and star ratings.