Collaborative Filtering Algorithm Based On Improved Time And User Impact

The prosperity of Internet technology has brought us a variety of Internet applications,which enrich our daily life all the time.It also causes all kinds of problems such as information overload.As one of the ways to solve the problem of information overload,recommendation system can predict the future behavior of users by mining their historical behavior and other records.But it will also face some problems and challenges.As for the existing collaborative filtering algorithms,they do not consider or seldom consider the influence of other factors on the recommendation,such as the time factor and the influence degree of different users.Some studies only consider it unilaterally,so the effect of user recommendation is often not as good as that of multi-factor consideration.Based on the traditional collaborative filtering algorithm and combining the above two factors,the work in this paper is as follows:1)A new incremental time weight function was proposed.Since most of the existing collaborative filtering algorithms generate recommendations based on users' historical score data or historical behaviors,they do not take into account that users' interests and hobbies may change,and their future interests and hobbies may not be consistent as in the past.First introduce the Ebbinghaus forgetting curve and analyze it to find the basic function expression that conforms to the changing rule of the forgetting curve,On the basis of keeping the original form unchanged,a new time weight function that conforms to the forgetting rule is proposed.When the user is predicted and scored,the scoring items that are closer to the current time occupy a larger weight.2)A new method to calculate the extent of user influence was proposed.Considering the influence of user attributes,the user influence was used as a factor to calculate the prediction score,and a new method to calculate the user influence was proposed.These turo factors make up the deficiency of traditional collaborative filtering algorithm that does not consider external influence factors.3)The improved algorithm(CF-IB)was tested in the Movielens dataset,and the experimental results were compared with the traditional collaborative filtering algorithm and the improved single algorithm of time weight and user impact in this paper,to verify the reliability of the algorithm.