A Subgraph Isomorphism Algorithm for Privacy Preserving in Dynamic Social Network

Abstract

Cyberspace topologies can be demonstrated mathematically by a graph theoretic approach which basic structure is in accord with the interconnected world of social networks. Subgraph pattern matching analysis is an approach to performing computer network data analysis that executes efficient continuous queries on dynamic social graphs and minimizes the overhead on both search space and query. Privacy preserving has become a forward-looking interest both in industry and in academia in recent years, especially for social media streams and cyber data sources are prominent examples of high throughput and dynamic graphs, which made us inundate with the stream of updates. Existing research work about anonymizing a social network is still an open question for research. It is because altering attributes of vertices and edges might affect the neighbourhood relationships between vertices; removing or adding vertices and edges might also influence other vertices and edges as well as the properties of the social network. As a matter of fact, measuring information loss in the re-construction of graph is also a problem of research. Therefore, the technical challenges of privacy security have aroused a lot of attention, including rational dataset anonymity strategy, subgraph matching method, user queries processing without compromising sensitive information and and adversary background capability modelling. As today's defenses on the network are fast becoming obsolete and dealing with problems causing by flooding activities and untrustworthy nodes in the network is urgent. To this end, this talk will complement the corresponding defects of existing methods as well as enhance overall security posture. More specifically, this research will investigate: (i) how to perform privacy preserving graph analytics on encrypted graphs, and (ii) how to ensure that the procedure of matching can be protected from privacy breach. In addition, we also focus on incremental processing algorithm that can support high data velocity social data graph with distributed environment. It is hoped that by doing so, it can alleviate the burden of searching (sub-)graph patterns. Finally, quantification analysis on information loss under different data and anonymous structural methods will be studied to find out the most suitable ones for privacy protection. The objective is to build a complete and strong framework aiming at dealing with not only scalability, confidentiality privacy leakage, and authenticity, but also fairness towards aborting behaviors, which can improve the work of privacy-preserving technique to achieve efficiency, safety, and convenience in practical application. To preserve nodes’ privacy, all identities are removed as shown in Figure 1(b). Unfortunately, if an adversary has acquired some knowledge about the neighbors of an individual, the privacy leakage may still happen. If an attacker knows that Dave has two close friends who they are mutual friends and has another three friends who do not know each other, the 1-neighborhood graph of Dave as shown in Figure l (d), so the node representing Dave can be correctly identified since no other vertices don't share the same 1-neighborhood graph in the released network. To protect the privacy network, the traditional method is to guarantee that any individual cannot be located uniquely in the anonymized social network with a probability higher than 1/k, where k is a user- specified parameter carrying the same structural pattern in the k-anonymity model. By adding a noise edge connecting Fred with Greg, the 1-neighborhood graph of every vertex in Figure (e) is not unique, which can ensure an adversary with the 1- neighborhood knowledge cannot target any individual from this anonymous graph with a confidence higher than 1. However, the existing work about anonymizing a social network is limited since altering labels of vertices and edges may affect the neighborhood relationships between vertices, and removing or adding vertices and edges may influence other vertices and edges as well as the properties of the Social network.