Trusted Bundle Recommendation Algorithm Based On Matrix Factorization

Traditional collaborative filtering recommendation algorithm, the recommendation algorithm based on user or the recommendation algorithm based on item both have the data sparse problem, if the data is huge the processing progress is too slow and the new user problem which have no behavioral data also have problem, for mining the long tail of goods is not sufficient, and if recommendation algorithm user in the data set which is noised, if do not preprocessing the data set, the algorithm will be unsufficient and will cause tremendous deviation prediction. At the same time, the traditional matrix factorization algorithm and the improved algorithm based on matrix factorization algorithms are faced with the lack of prediction accuracy, convergence speed is too slow, not considering the long tail distribution and hidden features of the data mining is inadequate.In view of the above problems, put forward a new recommendation algorithm, the algorithm considers the frequent degree long tailed distribution problems and items of difference, and carries on the thorough research on the relationship between mining user, user trust relationship from the user’s implicit relations.First of all, not only the traditional dominant relationship between users of research, and study on the intrinsic link between users, namely the user’s implicit association rules mining, and implicit trust between users, puts forward the goods implicit trust relationship model, the model is capable of mining inadequate or mining is not accurate enough problems led to solve using the dominant feature of the traditional the mining of user relationship.Secondly, by using the improved Apriori algorithm, TSApriori algorithm is proposed, by using the algorithm of mining frequent items to efficient, solves the traditional frequent item mining methods is not accurate and the problem of insufficient.Finally, in this paper put forward the Trusted Bundle Recommendation Method. Focused on the user and item angle and analysis the relationship between them, the experimental analysis of the association rules of various complex, concentrated in the Netflix data and MovieLens data showed that the algorithm has good prediction accuracy, mining the long tail mining and user trust relationship items were effectively, were analyzed and compared to experimental results the effectiveness of the algorithm, and stability is demonstrated.