A clustering based scalable hybrid approach for web page recommendation

The distribution of the number of items liked by users plays an important role in designing recommender systems. In case of implicit feedback we rarely get some clicking events compared to large item based e-commerce sites, where preference information is not so rare. In this paper we present a novel hybrid recommendation system based on clustering of items using co-occurrence information of pages and content information of pages. These two different types of clusters are used in a parametric form to get aggregated recommendations based on the available preference information of users. Our experimental results on Yahoo! Front Page “Today Module User Click Log” dataset show that the content based clusters plays an important role for users having very less preference information and also the clustering based hybrid approach gives better overall performance compared to other approaches. More-over, clustering of items gives a scalable implementation which minimizes the computational complexity.