Collaborative Filtering Recommendation Based On Two Step Model

Due to the explosive growth of information,it is difficult for uers to choose what they want from the overload information.Thus,recommendation systems which analyze or model using the historical behavior data of users attract the attention of many scholars.This can alleviate the problem of overload information to some extent.Recommendation algorithms are the key of the recommendation systems,collaborative filtering recommendation algorithm(CF)because of its high intelligibility and no need of semantic analysis becomes one of the most widely used recommendation algorithms.CF faces some problems such as data sparsity,the trade-off between accuracy and diversity,real-time and so on,overwhelmingly blocking its growth.In addition,Top-N recommendation is more suitable for the current requirements of users.Therefore,this thesis mainly focuses on solving the problems of CF to study its recommendation on Top-N and improve the recommendation effect.The main research work of this thesis is:(1)Recommendation based on bipartite network of two step model is put forward.Recommendation based on bipartite network only uses the single model of selection,ignoring the rating of users,based on the analysis of user behavior patterns,recommendation based on bipartite network of two step model is put forward.The method uses two step model rating prediction to combine network based inference with CF.Experiments on MoiveLens dataset show that NBIT can improve the accuracy of recommendation.(2)It proposes two step Top-N recommendation based on similarity of attribute weight.CF declines in quality because of extremely uneven and sparse rating data.In this case,this thesis proposes two step Top-N recommendation based on similarity of attribute weight.It calculates the similarity by using the attribute to expand the user's interest.Experiments on the MoiveLens dataset show that the proposed algorithm can improve the accuracy and diversity of Top-N recommendation.(3)Two step recommendation based on similarity of attribute weight is parallelized by Spark.In order to solve the real-time problem of CF,two step Top-N recommendation based on similarity of attribute weight is parallelized by Spark and improve the effectiveness of the recommendation algorithm by means of architecture.Experiments on MoiveLens dataset show that way can improve the speed of computation.