Research On Recommendation Algorithms Based On Probability Matrix Factorization Integrating Time Factor And Item Clustering

With the rise of Internet cloud computing,users are no longer worried about the lack of resources.Instead,the real problem is how to find the resources that match their interest.Users often get stuck in facing a huge amount of data.Therefore,the recommendation system has emerged in this very moment.It has become an urgent need that a recommendation system can not only customize for users but also save users' time.The personalized recommendation system saves users who are confused in massive data and pulls them out of the dilemma of not meeting their real needs,in the meantime also improves the utilization of information.The recommendation system has been successfully applied in many fields,such as e-commerce,video websites,major portals.Therefore,it is of great social and practical significance to improve and innovate the recommendation algorithm.The model-based collaborative filtering algorithm is one of the most widely used recommendation algorithms today.In addition,the probability matrix factorization model has a good recommendation effect.It learns the potential features of users or items,decomposes high-dimensional matrices into low-dimensional approximation matrices for recommendation,and can effectively process massive amounts of data.Existing collaborative filtering recommendation algorithms have some drawbacks,such as the data being too sparse and ignoring the changes of user interest over time.Aiming at the shortcomings of the above recommended algorithms,this paper proposes a probability matrix factorization recommendation algorithm based on fusion time factor and item clustering.The time decay function is used to mine the potential feature relationships among users and find the user set that is most similar to the target user.Item clustering is used to find potential feature relationships between items and find the item set that is most similar to the target item,and then combine it with the probability matrix factorization algorithm.For the sparsity of data,this paper proposes a filling model based on the weighted Slope One algorithm of the user's neighbors.The weighted Slope One algorithm model of the nearest neighbor of the fusion user is used to fill the rating matrix,and the screening of the nearest neighbor of the current target user is added while considering the inter-item deviation to ensure that only users who are similar to the current target user's preferences score their participation in the recommendation.The advantage is that the use of a small number of high-quality data helps target users to predict their preferences for the product,making the recommendation more accurate.By comparing and analyzing with other algorithms on the real movie dataset,we can see that the proposed algorithm has better recommendation effect,and the recommended list is more in line with the user's personal preferences and demands.The algorithm proposed in this paper can effectively solve the problem of data sparsity and cold start in the field of live video and movie recommendation compared with the traditional recommendation algorithm.It is also a good consideration for the situation that the recommendation result is not accurate as the user's interest preference changes over time.In the future work,we will further study and solve the impact of time factors in item relationships,and the impact of holidays and seasonal changes on user preferences.I believe that after continuous study and in-depth study,we will get better results.