Optimization Of Collaborative Filtering Algorithm Based On Time And Type Features

The information overload phenomenon in the present society has made the recommender system play an increasingly important role,and various recommendation algorithms have been continuously proposed.However,with the rapidly increasing amount and complexity of network information,the problems and deficiencies of the algorithms have gradually emerged.The issue of information expiration is another new challenge that the current recommendation algorithms face.At the same time,after the large-scale systems have produced a large number of items to be recommended,how to select the most suitable Top N items for users is also the direction in which the current recommender system needs to be optimized.For the issue of information expiration,the previous studies suggest that the information long time ago have less reference than the recent information for the recommendation at present.Therefore,the time forgetting function was introduced as the time weight to weaken the influence of the past information for current recommendation.However,only considering the information weight decay with time is incomplete,it ignores the effect that the user's impression will be strengthened of the information continuously attentioned,which will enhance the influence of past information.Therefore,we propose a matrix factorization based collaborative filtering algorithm that combines time and type weights.Firstly,the improved time weight with the information retention period was used in the matrix factorization model to improved traditional algorithm.Then,in order to solve the problem of information influence enhancement effect,we propose the type weight to modify the improved time weight based on above algorithm.And through experiments on the movie data set MovieLens,we compared our improved algorithm with the neighborhood-based time weighted collaborative filtering algorithm and the matrix factorization based collaborative filtering algorithm,and proves that our improved algorithm can achieve better accurate results,and it is applicable for systems that using ratings for recommendation,such as movie rating recommender system,music rating recommender system,book rating recommender system.In a recommender system that has a limited number of recommended items,if there are many high-rating items predicted by the recommended algorithm,and for the recommendation to be effective for the user,the probability that the user adopts the recommendations should be taken into consideration.So,we propose a type-sensitive filtering algorithm based on PageRank.This algorithm builded a network graph about the movie type based on the user's viewing data in the movie recommender system,and based on this carried out an improved PageRank iterative process,then we proposed the user-choice tendency for movie type,the movies with a high degree of tendency can be recommended to the user.And compare commonly used methods recommended by rating from high to low,our experiment proves that the user will preferentially choose the items after being filtered by the type-sensitive filtering algorithm.