Research On Cross-platform Entity Resolution Algorithms In Social Network

In recent years,with the rapid development of computer and Internet technology,data mining technology is widely used in various fields.In data mining,entity resolution technology plays an important role in eliminating redundant and linking records,and it is becoming an important step in data preprocessing.On the other hand,along with the rise and development of social networks,the application of entity resolution technology in social network is constantly being studied and discovered.The research of cross-platform entity resolution technology in social network can facilitate the realization of the personalized service in the social network,and it is also important to maintain the security of social network information.User data of social network has multiple components,including friendship,personal profile and so on,and the study of cross-platform entity resolution in social network,mainly focus on the characteristic of data.Common algorithms focus on one part of the data,so there are some shortcomings and drawbacks.Based on the analysis of each part of social network data,this thesis combines attribute and structure,and proposes structure-based attribute similarity extrema algorithm to solve the problem of cross-platform user entity resolution in social network.The algorithm based on the structural characteristics.With a little known information as the starting point and the help of the attribute characteristics,the algorithm performs multiple iterations,and finally achieves the task of user entity resolution.On the basis of the proposed algorithm,this thesis introduces some new concepts and definitions in the various steps of the algorithm,and makes a series of optimization.One of the optimization is the use of weighted extended attributes to calculate the similarity of attributes in order to reflect the similarity between users more accurately.The concept of structural compactness is introduced to choose the starting point of algorithm iteration.The concept of dynamic threshold is proposed to avoid a series of problems caused by fixed threshold.And the result tailoring is used to meet the needs of various scenes.After the proposed algorithm,this thesis uses manual data and the data collected from the real social network to analyze the performance of the algorithm.This thesis uses a public data set from Facebook,as well as real social network data from Sina Weibo and Renren that contains structural and attribute information,then a series of experiments are designed and executed.Through the comparison of the experimental results,this thesis analyzes the performance of the algorithm from many aspects,and proves that the algorithm has superior performance.The experimental results show that the proposed algorithm can effectively accomplish the task of cross-platform entity resolution in social network,and the optimization of each step has certain improvement on the performance of the algorithm.