Research On Key Technologies Of Privacy Protection In Location-Based Social Networks

With the rapid development and wide application of technologies such as mobile Internet,positioning technology,and smart terminal equipment,the integration between Location-Based Services(i.e.,LBS)and social networks(i.e.,Location-Based Social Networks,or LBSNs)is deepening.This provides us with more interesting,richer,more accurate,and more personalized service.Both online and offline,we can socialize,work,live,and entertain,which has revolutionized our way of life.Therefore,LBSNs have become an indispensable part of our lives.On the one hand,LBSNs provide users with services such as querying,publishing,and recommendations based on information such as users’ locations and preferences;on the other hand,LBSNs are ’honest and curious’,providing intelligent and personalized recommendation services by collecting,mining,and utilizing user-related service data.The application of LBSNs has brought us great convenience.However,the openness of the network,complexity of the environment and personnel,as well as limitations in the technology itself result in a series of privacy risks for user information involved in LBSNs application services due to excessive use and disclosure.Protecting the privacy of users while ensuring service quality and efficiency is a challenging problem.In response to this problem,scholars have conducted a series of studies and proposed numerous privacy protection schemes for various application scenarios,users,and privacy needs.Among them,scholars must either sacrifice a certain level of service quality and efficiency to improve user privacy,or compromise on a certain level of privacy to ensure better service and efficiency.There is no universal scheme for privacy protection.Ultimately,we must make a tradeoff between privacy protection,service quality,and efficiency.Therefore,we conduct research on different application scenarios and privacy protection technologies in LBSNs,taking into account various user privacy requirements.Our goal is to propose personalized privacy protection schemes that are suitable for different application scenarios and users,and to achieve comprehensive user privacy protection that can resist attacks throughout the entire process of Point of Interest(i.e.,POI)querying,publishing and recommendation.The main research work and innovations of this dissertation are as follows:(1)In the application of the location-based POI querying,nonnavigational nearest neighbor queries are characterized by submitting multiple query requests in the same location within a short period.The POI’s unique address has a hierarchical structure.The existing solutions do not take the above two factors into account,making them vulnerable to statistical inference attacks that can lead to privacy leakage such as user’s location,organization information,and query content.To address the aforementioned issues,we propose a Dummy Generation Scheme that considers the Hierarchical Structure of Addresses(DGS-HSA).DGS-HSA employs dummy to realize anonymous protection of user’s privacy information.When generating dummies,the historical locations are first divided into grids according to the organization;secondly,select k dummies that meet the degree of privacy protection<L(k,s),Q1>from multiple grids.This means that the k dummies are distributed across at least s organizations,and k query content contains at least l POI query categories.And to prevent privacy leakage caused by continuous query requests from the same user at the same location,DGS-HSA replaces the real user’s location with any non-duplicate historical location within the grid where the real user is located.Finally,DGS-HSA can protect user’s location,organization and query content while resisting Intersection Inference attack for Location and query Content(I2LC),Furthermore,a multi-objective optimization problem is solved to achieve a balance between privacy protection and system overhead.We evaluate DGS-HSA from both theoretical and experimental perspectives,and the results indicate that DGS-HSA achieves the expected degree of privacy protection and is capable of resisting statistical attacks.(2)In the application of the location-based POI review publishing,the platform maintains a business’s reputation by publishing users’ reviews of the business.However,each review typically includes a display name,avatar,comment content,etc.,and each business corresponds to a unique geographic location.Users actively publish reviews and thus actively disclose private information such as personal trajectories,behavior habits,and preferences,which are vulnerable to Statistical Inference Attack(SIA)and cross-platform Connecting User Identities Attack(CUIA).Existing solutions mainly protect user privacy through partial anonymity or suppression of publishing.However,these methods can reduce system utility and cannot completely avoid SIA and cross-platform CUIA.To solve the aforementioned issues,we have studied and improved the method for evaluating review usefulness and privacy protection mechanism,and proposed a Cross-platform Strong Privacy Protection Mechanism(CSPPM).Based on the consistency threshold between user ratings and a business reputation score,CSPPM anonymously publishes high usable reviews while filtering out false or factually inaccurate ones.While ensuring the utility of the system,CSPPM reduces excessive disclosure of user information and avoids the cross-platform CUIA caused by the display name.Furthermore,reviews are ranked based on the consistency between user ratings and business reputation scores to further enhance the usefulness of reviews.Finally,we assess CSPPM from both theoretical and experimental perspectives,with the results indicating that CSPPM is effective in resisting attack while maintaining system utility.(3)In the location-based POI recommendation scenario,data sparsity problem commonly exist,which affects the performance of the recommendation system.An effective solution to this problem is crossdomain location recommendation,which involves sharing user interaction data from the source domain with the target domain across domain in order to improve the recommendation effectiveness of the target domain.However,this method of cross-domain data sharing can easily result in the abuse of user data from the source domain by the target domain,thereby compromising user privacy.Therefore,we propose a Privacy-Preserving cross-domain personalized location Recommendation based on similar user Clustering(PPRC).On the one hand,PPRC maintains high availability of the data by sharing behavior patterns of similar user groups and achieves personalized recommendation.In the source domain,by mining user behavior patterns and clustering similar users,the behavior patterns of similar user groups that reflect user personalization characteristics,and POIs related to depersonalized similar user groups are shared with the target domain.In the target domain,the behavior pattern of the target user is extracted and matched with that of similar user group behavior pattern after fusing the source and target domain to achieve personalized POI recommendations.On the other hand,similar user clustering can remove the personalized characteristics of users,break the one-to-one relationship between‘user-POI’,and achieve protection of user privacy in the source domain.Finally,the recommendation performance,privacy protection effect and adaptability to sparse data of PPRC are evaluated from both theoretical and experimental perspectives.The results indicate that PPRC is effective.