Incorporating multidimensional information into dynamic recommendation process to cope with cold start and data sparsity problems

Area-of-Interest (AOI) recommendation is a type of context-aware recommendation that works based on location-based data. A context-aware recommender system should be able to provide the recommendations to the users based on their request time. Often, existing context-aware systems assume that the context information is constant over time and ignore dynamic users’ preferences. Besides, the development of a technique for incorporating auxiliary information within the recommendation models is a challenging task for the recommender systems. On another hand, the context pre/post filtering methods are inefficient in dealing with cold start and data sparsity problems. In these issues, there is not sufficient data to provide the accurate recommendation for the users. In this paper, a dynamic contextual modeling approach is proposed for improving the personalization in the sparse dataset. For this purpose, a density-based clustering algorithm is used for discovering the AOIs and coping with the sparsity problem. In addition, a hybrid similarity measurement is proposed to incorporate the auxiliary information into the recommendation process and suggest dynamic personalized AOIs in which the dynamic preferences of users are computed implicitly. The proposed similarity scheme using the auxiliary information can determine the neighborhood users for the cold start users, and accordingly, it can provide a list of recommendations to a target user. The experimental results based on Flickr and Yelp datasets demonstrate that the proposed method outperforms prior work on all three metrics, achieving a 10% increase on precision, a 25% increase on recall and 12% increase on F-Score in terms of quality metrics.