Why to ask one, when we can ask thousands? It's the motto of all
researchers, willing to put the task of Web Information Retrieval in the
sphere of Peer-to-Peer Technologies. Nevertheless, always, it's impossible
to actually ask thousands for results, so we must preliminarily limit the
number of "peers" to ask.
In our setting every peer has local index lists for the documents that it
has previously crawled. Certainly, we are able to choose best peer
collections, only having some summary about their content. These summaries
must be quite short for scalability. At the same time they must possess
enough discriminative power for estimation of response potential of a peer.
What summaries to choose, how they can be combined into one measure and how
we can use other evidences of user information need, for example, collection
of personal documents to improve our peer selection is the set of questions
being answered in the presented Master thesis project.