Research On Personalized Recommendation Algorithm Based On Transfer Learning

In the era of big data,the amount of information on the Internet is exploding,and the problem of information overload is becoming more and more serious,which has brought troubles to users.Therefore,the recommendation system comes into being,which can solve the problem of information overload to a certain extent.At present,the research on personalized recommendation algorithm has made a lot of achievements and has been applied in various fields.However,due to the sparsity of data in real scenes,the progress and development of recommendation algorithms are limited.In view of the wide application of the idea of transfer learning,some scholars have carried out relevant researches and proposed to apply the idea of transfer learning to the recommendation algorithm.The existing recommendation algorithm combined with the transfer learning idea can alleviate the problem of data sparsity and improve the accuracy of recommendation,but there are still the following problems: insufficient mining of the internal connection between auxiliary data and target data,insufficient information transfer and poor recommendation performance.To solve the above problems,this thesis carries out in-depth research and proposes two personalized recommendation algorithms based on transfer learning: transfer by collective matrix factorization recommendation algorithm considering nearest neighbors(TCMF-NN)and multi-source triple-bridge transfer recommendation algorithm(MSTBTR).Experiments on the relevant real data sets show that the two proposed algorithms have higher recommendation accuracy than the existing algorithms.The main work of this thesis is as follows:1.Deeply study the existing personalized recommendation algorithm and the personalized recommendation algorithm combined with the transfer learning idea,analyze and implement the corresponding algorithm,and find their the shortcomings.2.Combined with the idea of matrix factorization and transfer learning,transfer by collective matrix factorization recommendation algorithm considering nearest neighbors(TCMF-NN)is proposed.The existing transfer collective matrix factorization recommendation algorithm ignores the importance of neighbor users when performing collective matrix factorization on the user binary preference matrix and the user rating matrix.Therefore,TCMF-NN algorithm not only performs the transfer of the user's binary preference features and ratings features,but also considers the impact of neighboring users on user ratings,and completes the transfer of binary preference features and rating features of neighboring users,and integrates the rating features of the neighboring users into the user's rating prediction.Compared with existing algorithms,this algorithm can dig deeper into the internal connection between auxiliary data and target data,which makes the information transfer more sufficient and improves the recommendation accuracy.3.An auxiliary data construction algorithm BP-ADCA is proposed to construct binary preference auxiliary data from the original rating data for the transfer recommendation algorithm,which avoids the problem of recommendation performance degradation caused by the low correlation between auxiliary data and target rating data.Combined with the BP-ADCA algorithm and TCMF-NN algorithm,an improved transfer collective matrix factorization model is proposed.4.A new multi-source triple-bridge transfer recommendation algorithm(MSTBTR)is proposed to solve the problems such as insufficient information transfer,poor recommendation performance and the fact that most recommendation algorithms only use single source domain knowledge for transfer.Different from the existing recommendation algorithms combined with transfer learning idea which mostly adopt single bridge transfer patterns or build transfer method,the algorithm adds the user rating mode and the item rated mode to the twin-bridge transfer mode,and adopts the multiple source domain data transfer,and puts forward a way of multi-source triple bridge transfer,making information transfer more fully,improving the accuracy of recommendation.5.Experiments are carried out on several commonly used real recommendation datasets,and performance comparisons are made with existing algorithms to verify the feasibility and effectiveness of the algorithm proposed in this thesis.