enemos-p: An enhanced emotion specific prediction for recommender systems

Recommender systems suggest relevant item(s) to a new user by analyzing the existing user/item data. In recommender systems, collaborative filtering (CF) is a widely used technique to understand user preferences. The CF technique predicts a new user preferences to suggest the items by employing existing users’ preferences expressed in purchase history, ratings, and reviews. We observed from the real-world data that the user-rating data and user-review data given by a user are not expressing the same intent (we call inconsistency). This inconsistency in user-rating data and user-review data influences the performance of the recommender system. In this paper, we address the problem of inconsistency, and propose a novel approach that provide true recommendations. In this approach, we propose a refined approach called Enhanced Emotion Specific Prediction to refine the user-rating data considering emotional features in the user-review data. In the proposed approach, we exploit the multi-polarity to compute the emotional features from the emotional words. The emotional words are extracted at sentence-level by proposing sentence-level emotion detection algorithm. To validate the efficiency of the proposed approach, we conducted experiments on real-world data sets, Amazon and Yelp. We have used Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and Normalized RMSE (NRMSE) for comparing the performance. The results of the experiments show that the proposed approach significantly reduces MAE, MAPE, RMSE, and NRMSE compared to the existing approaches.