Research On Privacy Preserving Publication Algorithms And Recommendation Methods Of Social Networks

With the development of information technology, communication between people is increasingly electronic. At the same time, all kinds of Web2.0websites are highly popular, ways of online communication are further expanded, and the contents of individual creation are growingly diversified and enriched. These lead to the production and storage of large amounts of social network data every day in real life. Information-rich social network data has attracted the attention of both researchers and attackers. The challenge of how to publish social network in a matter that enables good utility and avoids privacy disclosure and of how to improve user involvement by recommendation so as to generate more valuable data has become an issue of great theoretical and practical significance.The publishing of social network data embodied by a graph with information whose nodes are with attributes may face the risk that an attacker identifies the user(node) via structural background knowledge or node attribute background knowledge, and the risk that the attacker does not identifies the user but acquires some sensitive information of the user. In order to protect user privacy, clustering of the nodes and edges of the original graph can be carried out so that each cluster contains multiple nodes. Meanwhile, node attribute generalization can be done thus avoiding attacks based on structural background knowledge or node attribute background knowledge. At the same time, the diversification of the generalized attributes of each cluster can be done during clustering so as to avoid the attracter from stealing the sensitive information of the user. To minimize information loss incurred by generalization and clustering, the greed strategy can be used, ensuring that an optimal node is added into each cluster each time. The final graph obtained via this way can effectively prevent the attacker from identifying the users and from obtaining a user’s sensitive information.The publishing of social network data, embodied by a graph whose nodes and edges are with attributes may face node disclosure, edge disclosure, and the disclosure of the sensitive information of the nodes or edges. If the sensitive information of a node or edge is disclosed, the sensitive information of the adjacent nodes or edges or associating edges may be identified. If the sensitive attributes of a node or edge is not identified, the sensitive attributes of the adjacent nodes or edges or associating edges cannot be disclosed. On this basis, in order to prevent an attacker from determining the values of the nodes or the path length between the nodes or the sensitive information of a node or an edge, the original graph could be divided into multiple discrete isomorphic connected subgraphs. Next, the corresponding nodes and edges among the subgraphs could be classified into multiple groups, respectively. Then, noise data is introduced to avoid the homogenization of the attributes of the same subgraph group. Finally, a connected subgraph and all node groups and edge groups are released. This way of publishing a graph can avoid node disclosure, edge disclosure, sensitivity disclosure of the nodes and edges, which can be proved by theoretical reasoning and be validated by experiments.Recommendation in social networks usually conforms to a variety of explicit or implicit criteria. Different specific criteria may have similar properties. When recommendations are evaluated in a social network, relevant subgraphs can be selected from a graph according to the recommendation criteria, or a number of graphs can be mapped into a relatively simpler graph, or original graphs can be simplified by setting node attributes as new nodes so as to construct a new graph. On this basis, the criteria can be classified by property and be processed respectively. Finally, the recommendation results from multiple criteria could be combined by a certain strategy. Experiments can verify the validity of the method.Personalized recommendation in social networks includes social data and personalized strategy. Statement of the recommendation strategy of the users’personal preferences is required not only to be easy to understand by both the users and developers but also to be able to guide computer unambiguous implementation. To meeting this requirement, a rule-based personalized-recommendation strategy can be developed. First, the users’preferences for recommendation strategies can be collected, obtaining commendation rules that are close to natural language but lack sufficient clarity. Then, each rule collected is reasoned into an unambiguous rule. On this basis, according to a certain strategy, eliminating the conflicts among the rules forms an unambiguous rule set. Finally, the rule set is used as a guide to measure the degree of recommendation among nodes. Assessment based on real data sets can verify the validity of this method.