Using Hierarchical Graph Maps to Explain Collaborative Filtering Recommendations

AbstractThe explanation of and justification for recommendation results are important objectives in recommender systems because such explanations and justifications strongly influence the user's trust in the system. Traditional justification methods are based on textual explanations, which can be inadequate for analysis, comprehension, and decision making on the part of the user. In this paper, we present a method that generates tree graphs that contain the following information: the recommended items, the items that have appeared most often in the recommendation process, the relative importance of the items, and the relationships that exist among the items. The trees obtained in the experiments show 1 the greater novelty of user-to-user results, 2 the overspecialization inherent in the item-to-item approach, and 3 the equilibrium obtained by employing hybrid user-to-user/item-to-item collaborative filtering. The proposed method presents the possibility of extending recommendation result justifications to groups of users and facilitates the explanation of large numbers of recommended items.