A recommendation system based on a lot of data generated through the development of digital technology is being introduced. Studies related to the existing recommendation system to provide customized information to individuals have been mainly analyzed using structured data that are easy to quantify, such as users' purchase status, ratings on items, and the number of visits. There is a problem that recommendation limited to structured data is less accurate. This paper aims to derive an emotional index by analyzing the review data left by the user, generate a new rating using the derived emotional index, and compare it with a recommendation system limited to the rating. Performance evaluation was conducted using user-based collaborative filtering, item-based collaborative filtering, and singular value decomposition method. Two directions of evaluation were conducted, the first using RMSE based on rating prediction, and the second using Precision@K and Recall@K based on ranking prediction. It was found that the performance of the recommendation system was better when using the corrected rating using the review data proposed in this paper in two directions: the rating prediction base and the ranking prediction base.