Abstract: In order to solve over-centralized distribution, low credibility and other problems in the rating of e-commerce websites,
this paper proposed an emotional analysis based on the content of comments. Namely, feature words and emotional words were
extracted to establish the feature vector of the comments and calculate the emotional tendency. On this basis, the original ratings were
corrected to obtain the rating better reflecting the real thoughts of users so as to improve the recommendation accuracy of the
recommendation system.

Key words: comments, emotional analysis, user rating, correction