An architecture for information retrieval over semi-collaborating Peer-to-Peer networks

Abstract

Peer-to-Peer (P2P) networking is aimed at exploiting the potential of widely distributed information pools and its effortless access and retrieval irrespectively of underlying networking protocols, operating systems or devices. However, prohibiting limitations have been identified and perhaps the most important one is the successful location of relevant information sources and the efficient query routing in large, highly distributed P2P networks. In this paper, a novel, cluster-based architecture for IR over P2P networks is presented and its evaluation is focused on retrieval effectiveness. We reason in favour of using clustering for P2P IR, by considering two fundamental hypotheses drawn from current P2P file-sharing systems. We also study the potential usefulness of a simplified version of Dempster-Shafer (D-S) theory of evidence combination for results fusion in the network. We simulated the IR behaviour of the system by using the TREC 6 and 7 ad-hoc track. The proposed architecture bears very promising results in terms of precision and recall.

Rewiring strategies for semantic overlay networks

“…In [13], Lu et al propose a two-tier architecture, according to which a peer provides content-based information about neighbouring peers and determines how to route queries in the network. Along the same lines, Klampanos et al [8] propose an architecture for IR-based clustering of peers, where a representative peer (hub) maintains information about all other hubs and is responsible for query routing. Loser et al [11] propose a three-layer organisation of peers (based both on peer content and usefulness estimators) and suggest combining information from all layers for routing queries.…”

“…Peer interests may vary from descriptions of document collections [13,20], to topics in a hierarchy [8], schema synopses [7], or ontology concepts [23]. In iCluster [20], the interests of a peer are identified by using its local document collection, and a single peer has a tunable and dynamic number of interests depending on its capabilities, collection size and content diversity.…”

“…Our architecture is general enough to cover fundamental design aspects of existing semantic overlays: as typically assumed in the literature peers are categorised by their content [8,10,13,14,20,23,25] and links to other peers involve not only similar peers, but also dissimilar ones to support the connectivity between different clusters as in [14,20,23,25]. The rewiring procedure is executed locally by each peer and aims at clustering peers with similar content, while queries are forwarded to peer clusters that are similar to the issued query.…”

“…To show proof of concept, we adopt a generic P2P architecture that utilises widely adopted techniques from SONs, such as characterisation of peers by content, peer clustering and link rewiring. In our SON architecture, peers are characterised by documents in the spirit of [8,10,13,14,20,23,25]. Each peer in the network maintains links to other peers with similar content to form clusters.…”

iCluster: A Self-organizing Overlay Network for P2P Information Retrieval

“…However, the work referred above assumes one interest per peer (peer specialization). Klampanos [11] proposed an approach for clustering peers holding information on more that one topics. Parreira [12] introduces the notion of "peer-to-peer dating" for allowing peers to decide which connections to create and which to avoid, based on various usefulness estimators.…”