Research And Implementation Of Malicious Users Detection Technology In Online Dating

Marriage social networks are increasingly popular which accompanied by emotional deception and money fraud.In order to ensure the safety of normal users in dating platform,it is important to identify and isolate malicious users in time.However,there is a lack of research literatures on malicious users detection in dating platform and existing literatures don't provide comprehensive analysis of malicious users.There are three problems with malicious users detection in dating platform:thefirst is malicious users personalization,the second is the imbalance of the normal/malicious users dataset;the third is the lack of users'researchable information.Based on the eXtreme Gradient Boosting(XGBoost)algorithm,this paper researches and designs a user trust model and new data balance methods to solve the above problems.The details are as follows:(1)User Trust Model is established to solve the personalization problem of malicious users.The structure of dating social network,users' characteristics and malicious users' types are studied.The users' basic information,behavior characteristics,interactive dialogue contents and image data are combined to represent the user trust model which can be used to find the difference between normal users and malicious users.(2)New data balancing methods are proposed to improve the recall rate of malicious users detection.In order to solve the mixed problem of the imbalance of normal/malicious users' dataset and incomplete labels of malicious users,the mixed sampling idea and the clustering algorithms are used which could solve the above problem.(3)Based on the user trust model and new balance methods,User Trust Model Balance XGBoost(UTMB_XGBoost)malicious users detection method is proposed.By comparing with commonly detection methods,it is found that the UTMB_XG Boost detection effect works best with 54.55%accuracy rate and 60%recall rate.(4)In order to optimize the lack of UTMB_XGBoost,Optimization UTMB_XGBoost(OUTMB_XGBoost)malicious users detection method is proposed to further improve detection accuracy and recall.It uses matrix decomposition method to solve the problem of lack of user researchable information.The BM25 and Word2Vec algorithms are used to improve the accuracy of users'conversation maliciousness detection.Compared with the UTMB_XGBoost method,the detection accuracy of the OUTMB_XGBoost method is increased by 4.61%to 59.16%and the recall rate increased by 13%to 73%.