A Collaborative Filtering Method for Personalized Preference-Based Service Recommendation

Abstract

Existing service recommendation methods, that employ memory-based collaborative filtering (CF) techniques, compute the similarity between users or items using nonfunctional attribute values obtained at service invocation. However, using these nonfunctional attribute values from invoked services alone in similarity computation for personalized service recommendation is not sufficient. This is because two users may invoke the same service, but their personalized preferences on nonfunctional attributes that describe the service may be different. Thus, to accurately personalize service recommendation, it is necessary for CF-based recommendation systems to incorporate users personalized preferences on nonfunctional attributes when recommending services to an active user. This paper proposes a CF-based service recommendation method that considers users' personalized preference on nonfunctional attributes. We first compute the satisfaction of an active user's preference on nonfunctional attribute(s) and then use these satisfaction values to obtain their similarity measures. We then employ the top-k algorithm to identify neighbors of the active user and subsequently, use the weighted average with mean offset method to predict his/her nonfunctional attribute. We evaluate our method using real-world services and also conduct experiments to show that the proposed method improves recommendation accuracy significantly.