Multi-view Clustering in Collaborative Filtering Based Rating Prediction
Ligaj Pradhan, Chengcui Zhang, Pradip Chitrakar · 2016
Collaborative filtering (CF) based rating prediction systems predict unknown ratings of a user for an item, based on the analysis made on how similar users rate similar items. Model based approaches such as cluster models provide efficient means to find out similar users or similar items as they inherently reduce the search space by previously grouping similar users or items into clusters. Hence, the accuracy of such cluster-based approaches can be improved by improving the clustering process itself. As such, in this paper, we present how multi-view clustering can be used to cluster users or items leveraging information from multiple modalities and improve the accuracy of CF-based rating prediction systems. We use Yelp business rating dataset to test our approach on user-restaurant rating prediction. We begin by identifying multiple views for both users and restaurants. Each view represents a distinct source of information regarding the user or the restaurant. These views are used to perform multi-view clustering for both users and restaurants, respectively. To predict the unknown rating of a user for a restaurant, we compute the averages of k-Nearest Neighbors (k-NN) from the respective user-cluster and the restaurant-cluster. Our results suggest that multi-view clustering is capable of leveraging the utility of each view to cluster similar users or items together and to improve the accuracy of CF-based rating prediction systems.