User Preference Multi-criteria Recommendations Using Neural Collaborative Filtering Methods

Abstract

Traditional recommendation systems work with a single rating provided by a user on an item. However, in many domains such as tourism, hotels, etc., a user would love to give rating for every criterion of an item based on his several experiences throughout the journey and at the same time, a user would love to be recommended considering all the features an item. Hence, the single rating recommendation systems are a bit inadequate for recommending items to the user in these situations and especially when user preferences change dynamically specifically on the criteria of the items that are to be recommended. So to tackle these anomalies in user behaviour, we propose a modified version of deep neural collaborative filtering method that is capable of predicting the criterion ratings of the items. To compare the similarity between criteria ratings and user’s dynamic shift in preferences on the criteria of items, we use some standard similarity techniques. The proposed network is designed to learn from user–item interaction data to predict the criterion ratings. To evaluate our proposed approach, we predict the overall rating of an item through using its criterion ratings with artificial neural networks (ANN). The proposed approach is tested on Yahoo! Movies dataset and TripAdvisor dataset. The proposed approach outperformed many of the baseline recommendation system models.