Query Routing in Peer-to-Peer Web Search
Pavel Serdyukov, Gerhard Weikum, Sebastian Michel · Max Planck Institute for Plasma Physics · 2005
The database selection task within a system of cooperative distributed libraries or web search engines has been thoroughly studied already for almost ten years. The most successful selection methods do not differ significantly from popular document ranking measures. They operate on the space of databases as virtual documents, composed of the text sources stored in databases. Web search traditionally deals with very short descriptions of information requests. This limitation leads to severe decreases in performance for all selection methods. The source of this weakness lies in the extreme ambiguity of short web queries with respect to enormously long virtual documents. In this work, we study the applicability of the authoritative database selection methods within MINERVA1, a peer-to-peer web search engine. For some methods we have made necessary system-specific adaptations in order to use them in a peer-to-peer environment. As our main contribution, we suggest using the context of web search to build a detailed description of a search request. We consider a personal document collection to be such a context and propose either to use the collection of the user who sends the search request, or to dynamically search the system for the collection capable to serve as a possible context. We propose the context-aware selection method and evaluate combinations of pseudo-relevance feedback methods within its scope. We demonstrate that our approach is more effective than the existing ones.