Research On Personalized Recommendation Algorithm

With the rapid development of computer technology and network technology,Internet applications and services have undergone explosive growth.Hundreds of millions of resource information have been generated."Information overload" has become an important issue that affects people's work and life.As one of the main ways to solve the information overload,the recommendation system has been widely used in many fields such as social network,content service and e-commerce.Although great progress has been made in the research and application of the recommended system,over time the external situation and internal demand have undergone tremendous changes.In order to adapt to the trend of technology development,this paper studies from two aspects: one is to study the sparseness,cold start and concept drift of the proposed system in the collaborative filtering algorithm;the second is to study the recommended sequence in the group recommendation scenario Fairness and sequencing issues.Papers mainly include:(1)A collaborative filtering algorithm based on neighborhood model and dynamic time is proposed.Firstly,a more accurate reference value preference model is used to improve the accuracy of the similarity.Secondly,the problem of the error in the score prediction algorithm is proposed.Then a calculation method combining the behavior data and the joint derivative interpolation weight is proposed.Then,in order to cope with the recommendation process The user's preferences change with the time characteristics of the introduction of the time attenuation function to improve the dynamic changes of the restrictions on the accuracy of the recommendations;Finally,the data is extremely sparse redefinition of the cold start problem,using the threshold conditions to judge the score results Fixed.(2)Based on the research of group recommendation system,a matrix decomposition recommendation algorithm based on group recommendation system is proposed.Firstly,the problem of fairness is improved from the perspective of group discovery,the SVD algorithm is used to identify the implicit semantic factors in the data,and then the implicit semantic factors are used to divide the user groups.Secondly,Recommended sequence sorting problem,using the algorithm of conditional walk two maps to sort the recommended sequence.(3)In order to highlight the effectiveness of our algorithm,used the common data set MovieLens to verify the above improvements.In the experiment of collaborative filtering algorithm,the recommended algorithm in this paper is 1.8% and 1.9% lower than the traditional algorithm in the RMSE and MAE respectively.When the data sparsity is 96.217%,97.478% and 98.739% respectively,compared with the other two And the RMSEs and MAEs are maintained at around 0.93 and 0.73.By comparing the effectiveness of the proposed algorithm,the proposed algorithm can improve the group recommendation results to the maximum extent and improve the effect when the group size is 64 Up to 3.91%.Experimental results show that:(1)The improved collaborative filtering recommendation algorithm is more accurate and can solve the sparsity,cold start and concept drift to some extent.(2)The group algorithm of clustering the implicit semantic factors by using SVD algorithm to identify implicit semantic factors can improve the satisfaction of group users with the recommended sequences.The conditional walking bipartite graph method can improve the recommended sequences,Improve the consensus of group users.