Development of recommendation systems using game theoretic techniques

Abstract

In the present work, we inquire the use of game theoretic techniques for the development of recommender systems. Initially, the interaction of the two aspects of the systems, query reformulation and relevance estimation, is modelled as a cooperative game where the two players have a common utility, to supply optimal recommendations, which they try to maximize. Based on this modelling, three basic recommendation methods are developed, namely collaborative filtering, content based filtering and demographic filtering. The different methods are then combined to create hybrid systems. In the weighted combination, the use of game theoretic techniques is extended, as it is modelled as a cooperative game. Finally, the methods are combined with the use of a genetic algorithm where game theory is used for the parent selection process. Our work offers a baseline for the efficient combination of recommendation methods through game theory and in addition the novelty method, Choice by Game, for the parent selection process in genetic algorithms which offers consistent performance improvements.