Research On Association Rule Recommender System Combining Time And Score Information

With the rapid development of Internet information technology and the launch of various service platforms,we are in the era of information explosion.It becomes more and more difficult for people to find the information they are interested in from the massive data.Therefore,the recommender system came into being.By analyzing the user's historical behavior,the recommender system can find out the user's personalized needs,and then recommend the content that the user may be interested in to help the user get a better user experience.Nowadays,recommender system has been widely used in movie and video recommendation,music radio,social network,personalized reading and other fields.At present,there are many mainstream recommendation algorithms.Among them,association rule recommendation is widely used in industry because it can mine valuable data from a large number of data and is simple and effective.Association rule recommendation faces two important problems:one is how to mine more valuable association rules in massive data accurately and efficiently,that is,how to generate strong association rules.Second,how to reflect the user's personalization and their interests.All of these require researchers to make a lot of attempts while learning from traditional methods.When we use Apriori algorithm to mine association rules,we often have the same rule antecedent but more than one recommendation.In other words,Apriori algorithm is used to mine many redundant association rules.In order to filter out these redundant association rules and improve the value of rules in association rule library.This thesis proposes a weighted association rule recommendation algorithm which combines time and score information.First of all,we describe the user's personalization and interest from the perspective of time and rating.From the point of view of time,the user's viewing data is sorted according to time to generate orderly user number,movie type and corresponding rating data.Time weight is introduced to add time weight to the user's rating on the sequence,so as to describe the importance of time.From the perspective of scoring,this paper filters out the low scoring and studies whether the scoring will affect the recommendation effect.Then,time and scoring weighting are introduced to get the weighted value of association rules.Then,the weights and association rules are combined to select association rules whose confidence and support meet the threshold to generate strong association rules.Finally,strong association rules are used to recommend users,so as to improve the efficiency of the recommender system.Secondly,we use k-means clustering to mine thesimilarity of movie types,preprocess the data,and then use association rules mining algorithm combining time and scoring information to mine association rules,so as to make recommendations.Finally,the collaborative filtering recommendation algorithm and the improved association rule recommendation algorithm are combined to form a combination algorithm,which can reduce the defects of a single algorithm.In this paper,movielens movie data set is used for validation experiments,and the experimental results of the data set are analyzed.An effective association rule recommendation algorithm is proposed.The mining rules are weighted to filter a large number of redundant and worthless rules,and finally the recommendation is completed.The algorithm proposed in this paper is compared with the existing association rule algorithm.The experimental results show that under various conditions,the association rule recommendation algorithm combined with time and scoring information can improve the accuracy and coverage of the recommendation.