Research On Service Recommendation And Privacy Preserving For Location-Based Social Networks

Recent years, lots of sensor-embedded smart mobile devices have been successfully applied to location monitoring to promote the development of Location-based Social Networks (LBSNs). Owners of smart phones can access Internet and use location-based applications ubiquitously to experience different kinds of services. In LBSNs, users can mutually share the interested geo-information to user social media by making use of existing localization technologies (e.g. GPS, Wi-Fi or RFID,etc.). With the advent of big data, large numbers of service have been sent to proxy server, in order to provide the service with different kinds of function or performance to users. Due to the limitation of traditional service acquisition scheme and communicational mode, on the one hand,it leads to service requester can not select the satisfied service rapidly and accurately. On the other hand, some new service can not be selected by suitable users because it lacks a certain service description information.Service recommendation system is able to actively push service to target users according to users’ historical record of visited locations. However,existing service recommendation methods lack of the consideration on users’ living habit and behavior preference, which results the personalized and satisfied service can not be recommended to target users.Furthermore, the problem of privacy leakage is inevitable when subscribers enjoy the great convenience provided by LBSNs. In LBSNs,users are encouraged to upload their true location information to server, in order to achieve the personalized service experience. However, attackers can steal the ture location data of victims by making use of the "honest and curious" characteristic of LBSNs server. Also, the sensitive information(e.g. life pattern, behavior preference or the locations to be visited in future,etc.) of victims may be inferred according to background knowledge,which is dangerous for the personal security of users.In this thesis, it focuses on the key technologies of "how to recommend the personalized service to users","how to protect the users’personal privacy" and "how to achieve personalized service recommendation under privacy preserving" over LBSNs,and make in-depth studies. The main work can be reflected by two aspects: service recommendation and privacy preserving, which includes four research contents: (1) similar user discovery in LBSNs, (2) POI service recommendation in LBSNs, (3) trajectory privacy preserving in LBSNs and (4) privacy preserving-based personalized service recommendation,which construct a basic research system on service recommendation and privacy preserving. The detailed research works are shown as following.(1) In the ascept of similar user discovery, existing similar user discovery methods lack of the consideration on user preference, which leads to the study on mobile trajectory pattern-based latent friend discovery in this thesis. Through analyzing the characteristic and distribution of original trajectory data, this thesis designs two clustering algorithms. In addation, a TF-IDF-based location classification method is proposed to generate the semantic information of locations and build the mobile trajectory pattern of users. Finally, the latent similar users can be discovered by considering the activity sequence and type popularity. This study effectively solves the problem of data sparsity and improves the users’quality of experience for LBSNs service. It provides a necessary supporting technology for personalized service recommendation over LBSNs in future.(2) In the ascept of POI service recommendation, existing POI service recommendation methods have the problems of location limitation and data sparsity, which leads to the study on pattern and preference-aware POI service recommendation in this thesis. The geographical location information can be transformed into semantical location information by making use of the relationship between users and POIs. In addation, the user preference model can be constructed by considering location popularity and user familiarity. This thesis proposes a pattern extraction algorithm to effectively extract the user movement pattern. By matching the movement pattern of each user, the suitable candidate service for target user can be mined. Finally, LBSNs server recommends the top-k candidate service according to the proposed scoring mechanism. This study utilizes the mobile trajectory description method to effectively reflect the interest and preference of users and improve the expandability of service system.It provides a beneficial solving idea for personalized POI service recommendation over LBSNs in future.(3) In the ascept of trajectory privacy preserving, existing trajectory preserving methods have the problems of sensitive information leakage,low data utility and low adaptivity, etc., which leads to the study on preference-aware trajectory privacy preserving in this thesis. The original points can be restructured into stay region and location region, in order to build the location anonymization space. In addation, a privacy risk rating method is proposed in this thesis. According to different location preference of users, different location anonymization scheme is utilized to generate the anonymous location. Finally, the anonymous trajectory sequence is generated by orderly linking the anonymous location. This study is able to improve the data utility under the condition that user personal privacy is not stoled by attackers. It provides an effectively solving scheme for self-adaptive privacy preserving over LBSNs in future.(4) In the ascept of privacy preserving-based personalized service recommendation, existing trajectory preserving methods lack of the consideration on the trade-off between personalized service recommendation and privacy preserving, which leads to the study on differential privacy preserving-based latent trajectory community discovery in this thesis. The original trajectory sequence can be divided into different trajectory segments by making use of trajectory partition technique. This thesis designs location obfuscation matrix and trajectory sequence function to obfuscate the original locations and trajectory segments. During the stages of location obfuscation matrix generation and trajectory sequence function generation, Laplace distribution-based noise and exponential distribution-based noise are added to the outputs,respectively, in order to make the proposed optimal obfuscated trajectory sequence selection algorithm satisfy the ∈-differential privacy. Finally,LBSNs server receives the obfuscated trajectory sequences, it clusters the trajectory sequences into different kinds of communities according to geographical distance and semantical distance, in order to recommend personalized service for target users. This study effectively balances the personalized service recommendation and privacy preserving. It provides a feasible technical proposal for privacy preserving-based service recommendation over LBSNs in future.In summary, this thesis focuses on service recommendation and privacy preserving problems in LBSNs, and makes a systematic study from service recommendation, privacy preserving and privacy preserving guaranteed service recommendation, in order to bulid a whole research system, and output some basic research achievements. In study methods,this thesis adopts the methods such as related work survey, mathematical modeling and algorithm design etc. In data analysis, this thesis verifies the superiority of proposed methods by considering the actual performance index of service recommendation and privacy preserving in LBSNs. The research works in this thesis will have certain reference meanings for the study and development of LBSNs in future.