Research On Key Techniques Of Privacy Preservation In Personalized Search

As Internet grows fast, the information on the internet becomes explosively increasing. Accordingly, results from search engine also grow bigger and become harder for users to locate pages that they concern about. Personalized search is a solution to this problem. By collecting the user's information and analyzing what is his/her favorite topic, the personalized search engine can build up a user model to express the user's interest. When a search query is submitted, the result pages can be compared with the user model, and pages that match users' interest can be picked up and displayed to users, thus a personalized search is completed. Although this technology brings great convenience to users, it also raises a great challenge: the privacy issue. As the personalized search requires the personal information from users, users become worried about the leak of their privacy information. Some search service providers provides privacy policies to protect users' privacy, but it is still short of technical methods. This thesis mainly focuses on this problem and tries to provide some technical solutions for it.Related WorksFirstly, related areas of personalized search, privacy preservation and autonomic computing are thoroughly investigated. Current research issues and methodologies in these areas are analyzed. Also, the major advantages and problems of current techniques are identified, which arouse following research in this thesis.Autonomic Computing Based Privacy Preservation FrameworkThis thesis proposes a privacy preservation framework in personalized search by analyzing the basic requirements and principles in privacy preservation, in order to protect users' privacy in all the procedures in personalized search such as collecting, transferring, using and storage. Furthermore, an autonomic computing model is proposed to provide the self-configuration, self-healing, self-optimization and self-protection features.Cuttable User Profile Modeling Method To enable a user to control his/her personal information in personalized search, this thesis provides a cuttable user model built by the ODP and user's search records. By using the basic structure of ODP, updating the model with users' search keyword and page browsing and adding weight to user's different interests, user's personalized favorite topics can be clearly expressed. And by cutting the branch of the user model, the user's privacy information published to search engine can be controlled by user himself/herself.Mixed Page Sorting MethodUsing the cuttable user model, this thesis proposes a mixed page sorting method by using cutted user model in server-side and complete user model in client-side. The search engine server only holds the cutted user model and has the result pages sorted for the first round, while the client holds the complete user model and has the result pages sorted finally. Because the server only has a cutted user model, it cannot harm the user even it is leaked, since the user have deleted the information that he/she is sensitive.State Transfer Based Dynamic Access Control MethodFinally this thesis proposes a state-transfer-based dynamic access control model and offers the implementary architecture to protect the users' profiles. By monitoring the state change of all elements in a RABC system, and combining the new state and the old state as a state-transfer, policies can be defined and applied accordingly. This method is able to offer more appropriate policies for a changing system.