Research On Label-based Expert Information Recommendation System

With the development of big data application technology,the recommendation system has a wide range of applications.Douban online recommendation system will recommend books and movies for you because of your preference;On Netease Cloud Music,it will recommend similar music you have heard recently.The core of the recommendation system is the recommendation algorithm.Currently,the recommended algorithms are mostly based on collaborative filtering or content-based,and in some practical applications,tag data is needed.Tag data is the user's annotation of the content The recommended algorithm for processing tag data is temporarily no other two algorithms.In order to solve the problems existing in the existing application of the label recommendation algorithm,this paper studies in the following aspects:Firstly,an undirected weighted graph model is proposed for the existing problems of data labeling in the existing label-based recommendation algorithm.In the model,vertices represent data objects,edges represent behavior between data objects,and vertex attributes represent features of data objects.Based on this model,this paper proposes the VertexRank algorithm.Secondly,for the graph model,only the behavior of labeling between data objects is recorded,and a model for recording the number of occurrences of labels between data objects is proposed.Based on this model,the recommendation algorilhm TagBasedRank is designed to calculate the recommendation list in descending order by calculating the product of the number of times the data object administrator tags and the number of times the data object expert is tagged.The recommended list of the two medels is weighted and combined to obtain the final recommendation algorithm CombinationRank.Then,since the data processed by the above recommendation algorithm is data obj ect behavior data,not the data object itself.This can result in data objects that do not have behavioral data that cannot be recommended In view of the "cold start" problem of data objects,the author adds the similarity between the data objects to the above model,and then modifies the graph model and the label-based model to alleviate this "cold start" problem.Finally,the recommended effect of the Combination Rank algorithm proposed in this paper is verified by experiments.By comparing the recommendation algorithms before and after the improvement,it shows that the improved algorithm has belter performance than the unimproved recommendation algorithm in the recommended accuracy and coverage.The improved algorithm can better meet the needs of the actual application mthe recommended funcdon.Figure[21]table[8]reference[41]...