Research On Dynamic Collaborative Filtering Recommendation Based On User Preference Modeling In Recommendation System

In the era of big data,the data information is exploding exponentially.It is difficult for users to extract content that is useful or interesting for them from a huge amount of data.How to solve information overload is an important problem in artificial intelligence and big data age.The emergence of recommendation system becomes the breakthrough to solve the problem of information overload.It can help explore the potential interests of users and is an important branch in the field of personalized service.In the process of research on recommendation system,collaborative filtering algorithm is still the most widely used recommendation algorithm,whose core idea is to use the historical behavior data to mine certain similarity to recommend.but collaborative filtering algorithm also has some shortages,such as cold start,poor data sparseness,lack of accuracy and other key issues.On the basis of analyzing many recommendations and for the problems existing in the traditional collaborative filtering recommendation,this paper proposes a new dynamic collaborative filtering recommendation method based on user preference modeling,which is the improvement on the item-based collaborative filtering algorithm.Its main idea is that we use the Pearson correlation coefficient method for calculating the similarity between users and cluster users which makes the users in the same class have higher similarity and users in the different classes have lower similarity.After clustering,according to concentration of users' rating data in different classes,we determine the preferred items of users in the different classes and find more candidate neighbor items,which have more common scores with the target item.Then this paper generates the nearest neighbors of the target item after further screening.In addition,the time weight factor is introduced into the item similarity calculation and we determine the user interest weighted function by users visiting items in a certain period of time.The user interest weighted function reflects the change of users' interest in items with the change of time.The addition of time weight factor improves the accuracy of item similarity calculation and improves the recommendation quality further.Experiments are conducted based on the MovieLens data set,and the result shows that compared with the traditional collaborative filtering algorithms,the proposed collaborative filtering algorithm has better recommendation.