hide
Free keywords:
-
Abstract:
We consider a network of autonomous peers forming a logically global but
physically distributed search engine, where every peer has its own local
collection generated by independently crawling the web. A challenging task in
such systems is to efficiently route user queries to peers that can deliver
high quality results and be able to rank these returned results, thus
satisfying the users' information need. However, the problem inherent with this
scenario is selecting a few promising
peers out of an a priori unlimited number of peers. In recent research a rather
strict notion of semantic overlay networks has been established. In most
approaches, peers are squeezed into
a semantic profile by clustering them based on their contents. In the spirit of
the natural notion of autonomous peers
participating in a P2P system, our strategy creates semantic overlay networks
based on the notion of ``peer-to-peer dating'': Peers are free to decide which
connections they create
and which they want to avoid based on various usefulness estimators. The
proposed techniques can be easily integrated into existing systems as they
require only small additional bandwidth consumption as most messages can be
piggybacked onto established communication. We show how we can greatly benefit
from these additional semantic relations during query routing in search
engines, such as MINERVA, and in the JXP algorithm, which computes the PageRank
authority measure in a completely decentralized manner.