Research On Privacy Metrics In Online Social Networks

With the rapid development of the Internet in recent years,the cost for people to use the Internet is getting lower and lower,and now it has entered an era where everyone is a netizen.People publish and obtain information through various ways on the Internet.Among them,social networks have become the most important channel.As more and more users with different identities and backgrounds join in,problems also arise:privacy disclosure on social networks have become more frequent and serious,and security incidents caused by data leakage and abuse emerged in endlessly.What is more worrying is that these security incidents are user's personal privacy issues,and the consequences are usually small and far-reaching,so they have not been paid attention to by the users and researchers.However,the endless privacy leakage incidents has proved that there are huge loopholes in the personal privacy protection on social networks.The leakage of these information may not directly harm users,but will bring subsequent malicious behaviors such as network fraud,identity forgery and spam.At the same time,the large number of nodes in the social network and the frequent interaction between the users have caused great difficulties for privacy protection,and privacy protection solutions in related fields such as the Internet of Things and,storage networks are not applicable to social networks.Therefore,it is urgent to propose a method for personal privacy protection for the specific scenario of social networks to solve the above-mentioned problems,and privacy measurement have emerged.Based on the in-depth analysis of the characteristics of social networks and the summary of the deficiencies of existing research methods,this paper proposes a new privacy measurement method.The research contents are as follows:1.In view of the existence of multiple specific purpose social networks,different social sites have different backgrounds and privacy protection methods.Users may use multiple social networks at the same time according to their needs and existing methods cannot effectively protect user privacy.This paper proposes a method suitable for cross platform measurement of users' personal privacy,which quantifies the attribute information in each social network.We improve the accessibility,extraction difficulty,reliability and other indicators of the traditional methods,propose a new indicator of privacy awareness,and then get the visibility of attributes through the algorithm.Finally,combining the attribute sensitivity,we get the user's privacy score on multiple social networking sites.Experiments show that the algorithm can effectively quantify the privacy state of users through this score on multiple social networks.2.Homogeneity in social networks refers to that the more closely linked nodes are,the more similar their features are.This feature has prompted researchers to abandon quantifying the privacy status of users only from a personal perspective.Consider privacy leakage in the relationship:the more friend relationships around the user,the more complicated the network structure,the greater the possibility of inferring the privacy information of the target user by analyzing the privacy information of these friends.But the analysis of these friends faces a huge problem:social networks conform to the characteristics of small world,that is,anyone can reach each other within six hops,which leads to a large number of related users around targeted users,so that the existing algorithm cannot complete the analysis.In this research,we find that only a small part of these huge number of friends have a real impact on the privacy disclosure of the target users,and most of the others are redundant users.In view of the above situation,this paper proposes an improved algorithm to get the structural similarity between users and any friends.After sorting the structural similarity,the lower ranked users are excluded,that is,the friends with low relevance to the privacy disclosure of the target users.Then combined with the previous personal privacy measurement method,we can measure the user's privacy status in the entire network graph structure.3.In the above research,this paper found that while using structural similarity to screen redundant users,there is a lack of information timeliness,which exist the delay in the time.These related users may have a close relationship with the target user in the previous environment,but the information they hold at present has lost its privacy effect.This is because the target user's attribute information also changes with the environment,and some users who are out of the target user's living environment have not mastered these new information.Therefore,the information they have mastered has lost timeliness,and the disclosure of these information has no impact on the privacy status of the target user,which is an issue that has not been considered in the existing research.To solve this problem,this paper uses the behavior characteristics of the target user and their friends to propose the concept of behavioral intimacy,then combined with the structural similarity proposed in the previous section to obtain friends who closely related to the current privacy leakage of the target user and calculate the privacy score.Experiments show that this method can effectively solve the problem of lack of timeliness,so as to obtain more accurate privacy metrics.4.In the traditional research field of privacy measurement,researchers use to quantify the factors that affect the privacy disclosure of users,and finally get an overall measurement.In this process,it not only choose these factors artificially,but also ignore the implicit relationship between these privacies.At the same time,when analyzing the whole network,these traditional methods can only get one user's privacy score at one time,which is very inefficient.Therefore,this paper propose a framework to measure user privacy through deep learning model,which combines the factors of friend relationship,attribute information,behavior characteristics and so on.The framework can effectively extract the hidden relationships between these features.Meanwhile,it can get rid of the cumbersome feature selection and calculation steps.Experiments show that the graph neural network framework designed in this paper can efficiently and accurately obtain the privacy metrics of all users in the entire network.Based on the in-depth analysis of the results of the existing privacy measurement field,the above research results have proposed a privacy measurement method for multiple social network platforms and a user location for the situation of large user base,complex components,and complex network structure in social networks.The privacy measurement method in the network environment,the privacy measurement method for the user's behavior characteristics,and the privacy measurement framework using the deep learning model,these achievements have certain theoretical significance and practical score for the privacy protection of users in social networks.