Research On Conventional And Long-tail Binding Recommendation Based On Matrix Factorization Using Game Theory

In recommendation system, items tend to be long-tail distribution, on the one hand, a small number of items attracted most attention, and most recommended items from system tend to concentrate on a small number of popular items, formed recommendation in conventional sense; on the other hand, the huge number of long-tail items makes the users’actions of long-tail items often account for more than half of the total actions, it’s of great importance to recommend long-tail items to users. However, in recommendation system, most of recommendations belong to conventional recommendation, and even if there are a small part of the recommendations that considered long-tail recommendation, these recommendations did not considered the reasonable allocation problem between conventional recommendation and long-tail recommendation,To solve the above problem, conventional and long-tail binding recommendation that based on game idea was proposed. Firstly, the binding recommendation of conventional recommendation and long-tail recommendation was analyzed based on cooperative game, and the game model of conventional recommendation and long-tail recommendation was built. Then, tow kinds of methods that realize the proposed game model was proposed:(1) based on the idea of similarity constraint, the models of similarity regularization probabilistic matrix factorization were built and the modeling of conventional and long-tail’s binding recommendation was realized relying on the technology of matrix factorization.; (2) furthermore, the effectiveness and rationality of similarity constraint implementation method was discussed, and the model of adaptive similarity regularization probabilistic matrix factorization was proposed. The experiments of the tow game models were done on the data set of MovieLens 1M and Netflix, and the results show that the realizing method of adaptive similarity regularization probabilistic matrix factorization is better.Experiments also verified the effectiveness of the conventional and long-tail’s binding recommendation model, as well as the rationality of the cooperative game of conventional recommendation and long-tail recommendation in binding recommendation, and get the following important conclusions:(1) Conventional and long-tail binding recommendation problem can be solved by the idea of the theory game; (2) The balance of conventional recommendation and long-tail recommendation in their cooperative game can improve recommend effect in recommend model; (3) In the conventional and long-tail binding recommendation, conventional recommendation has more contribution than long-tail recommendation.