Client-side web mining for community formation in peer-to-peer environments

Abstract

In this paper we present a framework for self-formation of interests-based Peer-to-Peer communities using client-side web browsing history. We propose an order statistics-based algorithm to build communities with hierarchical structures. We also carefully consider the privacy concerns of the peer and adopt cryptographic protocols to measure similarity between peers without disclosing their personal profiles. We have evaluated our algorithm using a distributed data mining toolkit. The experimental results show that our framework could efficiently build interests-based communities.

Privacy-Preserving Data Mining in Peer to Peer Networks

“…Taking this into account Liu et al (2006) used cryptographic secure inner product protocols to compute the inner product between two users profile vectors. This was believed to have helped preserve user"s private data; however there was still room for improvement.…”

“…To explain this, there is no control over the user"s behavior and are also not monitored. The points put forward by Liu et al (2006) are seen to be computationally very intensive [7], expensive and not scalable in any way.…”

Peer-to-Peer Data Mining, Privacy Issues, and Games

“…We discuss an exciting application of web usage mining using the concepts of client-side web cache and P2P technology based on [13] and emphasize the need for privacy in such mining operation.…”

“…For any user it is imperative that its browsing data is not revealed in its raw form while forming these communities -otherwise it is almost impossible to convince web users to take part in such P2P computation where every user shares some data, does some computation and finally gets benefitted from the aggregate result (by being part of a similar minded community). In order to safeguard each user's private data, Liu et al [13] have used cryptographic secure inner product protocols to compute the inner product between two users' profile vectors. However, such secure multi-party protocols are based on honest/semi-honest third party assumptions, which assumes that a user or a group of users will follow the protocol as specified and will not form a malicious liaison or do anything to extract private information from other users.…”

“…In real-life, however, such ideal assumptions fall apart since very little control exists on each user's behavior and there is no centralized coordinator or administrator to monitor and govern all user activity. Besides, experimental results reported by Liu et al [13] show that this secure protocol is (1) computationally very intensive (2) ex-pensive from communication point of view and (3) not scalable at all. It is evident from the results that privacy preserving techniques designed for standard data mining is not going to work well in a P2P setup.…”