Research And Application On Mobile Context-Aware Recommender System

Recommender system is an effective way to solve the problem of information overload.Traditional recommender system only considers the relation between user and item,and taking no consideration of the context when user making decision.Context-aware recommender system(CARS)is the state of art recommender system in recent years.It can improve the recommendation accuracy by taking the context information into account while making recommendation.With the popularity of mobile devices,mobile devices give us massive information and bring mobile users serious information overload problem.The context information will affect a lot when users make decisions on mobile devices.Research shows that if CARS can use the context information properly,the recommendation accuracy can be improved.On the other hand,taking into account the context information can bring a series of challenges,in addition to increase the dimension of the data but also increase the sparsity of data.Moreover,the selection of mobile context information is also an important problem that mobile recommender system should pay more attention to.This thesis focus on some key issues that mobile context-aware recommender system should solved.First of all,according to the problem of context selection,we propose a context selection algorithm based on the relative standard deviation,and use an improved context-aware matrix factorization algorithm called CAMFB(Context-Aware Matrix Factorization with Bias)to verify the validation of context selection algorithm.Secondly,in order to solve the problem of the increasing dimension and sparsity of data,We propose a algorithm using higher order singular value decomposition also know as tensor factorization based on context weight,called WCATF(Weight-based Context-Aware Tensor Factorization).Finally,experimental evaluation on the real world dataset and semi-simulated MovieLens dataset demonstrates the efficacy of our algorithm.