Research On Related Issues Of Personalized Recommendation Strategy Based On Tag Spectrum Clustering

The prosperous development of Internet has brought far-reaching impact on modern society.Especially in the era of web2.0,people can carry out a series of activities on the Internet,such as online shopping,online office and online learning.People’s lives are filled with of documents,pictures,audio,video and other diversified Internet information.Both the producer and consumer of information are confronted by enormous challenges.Information overloading has become a problem to be solved.As a result,a personalized recommendation system is brought forward.To date,a large number of recommendation algorithms have emerged in the recommendation system field.In this paper,we discuss deeply about personalized recommendation system,and try to propose a personalized recommendation algorithm based on label spectrum clustering in terms of data sparsity,recommendation accuracy and so on.The main contents are as follows:First of all,this paper studies the basic principles and related background knowledge of several common recommendation algorithms.Among them,the advantages and disadvantages of neighborhood-based collaborative filtering recommendation methods and their respective applicable scenarios are analyzed and compared emphatically.Then,aiming at the common problems in recommendation system,combined with the advantages of labels and spectral clustering,a tag spectrum clustering collaborative filtering recommendation(TSCF)algorithm is putted forward.This recommendation algorithm has the following two characteristics:(1)With the increase of users’ awareness of personal privacy security,the demographic information that can be used in the traditional recommendation system is limited,and the bottleneck of similarity accuracy will limit the further improvement of recommendation quality.In this paper,UGC tags are used to mine users’ interests.The user relevance is used to modify the user’s similarity based,thereby improving the final recommendation accuracy.(2)By means of co-occurrence similarity of tags,modularity functions are designed and the tags are clustered by spectrum cluster algorithm.Based on the tag clusters,users with higher correlations are assigned to corresponding groups,and recommendations are conducted within the user groups.It can reduce the sparseness of data,as well as the search space of neighbors,and improve thereal-time and diversity of recommendations.To a certain extent,it can also ease the scalability of the recommendation system..Finally,using the data set published by the Lastfm Music website at the Fifth Rec Sys Conference,some simulation experiments were conducted to verify the validity of the proposed algorithm.By comparing with user-based collaborative filtering and recommendation based on user K-means clustering.Experiments show that the TSCF algorithm has improved in accuracy,recall and other metrics.