A Proposed Framework for Recommendations Aggregation in Context Aware Recommender Systems

The contextual information has been proved to be very useful and influential aspect by the researchers and domain experts in the recommender systems to improve accuracy while making recommendations. Generally, there are not enough preferences in multidimensional contexts or the preference matrices are sparse. To handle data sparsity problem and bring contextual effects, it is important to include valid and influential context features. To do so, this proposal first acquire most important and promising contextual information belonging to three different categories to drive data selection or data construction. Then, ratings are predicted on the selected data using two different approaches i.e. collaborative filtering on user based and item based model. Afterwards, a strategy is used to combine recommendations obtained from each category of contextual information. We performed experiments on movie recommendation and music recommendation domain. The results of our experiments show the advantages of this proposal over state-of-the-art methodologies.