The Research Of Personalized Recommendation Algorithm Based On Latent Factor Model

In the era of big data,various kinds of information have exploded.In the context of data overload,it becomes difficult for users to find the information they need quickly.Nevertheless,recommender systems can solve this problem well.Recommender systems provide personalized recommendation for users according to their behavior.Recommender systems have been widely used in major websites,which is of great commercial value.And,recommender systems are closely tied up with big data and artificial intelligence,which is valuable in academic research.Therefore,further research on recommender systems is of great significance.However,there are some critical problems in recommender systems,such as data sparsity,system scalability and so on.At the same time,the accuracy of recommendation needs to be improved.These problems are the main constraints on development of recommender systems.In order to alleviate the problems mentioned above,this thesis concentrates on recommendation algorithm.The main works and contributions of this thesis are as follows:(1)Due to the lack of neighbor data,the accuracy of the algorithm decreases dramatically.The recommender systems also face the problems of sparse data.A new recommendation algorithm by integrating latent factor model and potential classification correlation is proposed.The improved algorithm combines the latent factor model with the item-based collaborative filtering recommendation algorithm.The innovation of the proposed algorithm lies in calculating the similarity between items by using the potential classification correlation,and brings the time context into the prediction rating.Finally,an experimental analysis is carried out on the open data sets.(2)High-dimensional rating matrix consumes enormous computing resources,which brings about poor system scalability.A new recommendation algorithm exploiting latent factor model and gaussian mixture model is proposed.The improved algorithm combines the latent factor model with gaussian mixture model.In the proposed algorithm,the rating data is preprocessed by gaussian mixture clustering and divided into several smaller sub-matrices before recommendation,which reduces the complexity of computation.The influence of implicit feedback on user decision-making is also considered.Finally,an experimental analysis is carried out on the open data sets.(3)The performance of the proposed algorithms is tested on the intelligent community data platform of Chongqing Huiju Intelligent Electronics Co.Ltd.The results show that the optimal accuracy,recall rate and F1 measure of the latent factor model recommendation algorithm combining with potential classification correlation are 43.6%,42.8% and 40.2% respectively.The optimal accuracy,recall rate and F1 measure of the latent factor model recommendation algorithm combining with gaussian mixture model are 46.6%,45.8% and 41.6% respectively,and the memory usage of the algorithm is reduced by 24.97%.Both proposed algorithms can alleviate the problems of recommender systems effectively and improve the accuracy of recommendation.