Gravitation-based personalized recommendation algorithm

Abstract

A recommendation algorithm is proposed by introducing the universal law of gravitation into a recommendation system. This new algorithm is named as the gravitation-based personalized recommendation (GBPR) algorithm. In the algorithm, social tags used by users are regarded as particles that made up of their preference objects, social tags marking on items are considered as parti-cles that made up of item objects, and the user preference objects and item objects are taken as a user preference object model and an item object model, respectively. Gravitation exists between the user preference objects and item objects, and its strength obeys the universal law of gravitation. The strength of gravitation between the user preference objects and the item objects is computed, and it is regarded as their similarity. The bigger the strength is, the more similar they are, and the corresponding item objects are more proba-ble to be liked by users. Experimental results show that the proposed algorithm can get good performance.