| Abstract |
I present a new generation of peer-to-peer search systems inspired by BitTorrent. In our peer-to-peer search approach, each peer is both a search client (a system that submits queries to the network) and a search server (a system that answers queries). Peers that submit a lot of queries, will have to answer a lot of queries as well. Peers may answer queries by providing their own indexed collection, or by caching search results. I present simulations that show the effects of search result caching on query load balancing. I also show how the network can learn from the search result snippets without downloading or crawling the documents themselves. Finally, I will show an approach to include results from structured databases and other deep web search sites into the peer-to-peer search network. |