Research And Improvement Of Collaborative Filtering

With the arrival of big data era,the issue of information overload is becoming more and more serious.As an important means of information filtering,recommender system is one of the leading methods to solve the information overload issue.Collaborative filtering,without requiring content analysis and sharing the user experience information,has been one of the earliest,most wisely use and most successful techniques for recommender system.However,there is still room for further improvement in the performance of existing collaborative filtering techniques.In this context,this paper conducts further studies on collaborative filtering.Firstly,Gaussian iteration algorithm is proposed to solve the problem of slow convergence in the training process of three typical collaborative filtering models: neighbor model,latent factor model and hybrid model.By changing the objective function of the model,the model can adaptively adjust the gradient size in the iterative calculation process.In order to verify the effectiveness of the proposed algorithm,this paper applies the algorithm to three mainstream models of collaborative filtering,and carries out simulation experiments on public data sets.Experimental results show that the collaborative filtering models based on Gaussian iteration can effectively reduce the number of computing iteration and decrease the training time.Secondly,in order to solve data sparsity problem and cold start problem of collaborative filtering,based on the limitation of poor prediction accuracy in the leading latent factor model,a novel latent factor model with temporal dynamics is proposed to improve the prediction performance.The novel model decomposes the characteristic factors of users and items into concrete time-node factors,according to the users' rating time.The novel model is experimentally verified on the open datasets.The experimental results show that the new model can reduce the estimation error even on the extremely sparse data sets.