I have implemented (more or less) Rochio rel. feedback. You have to make some minor modifications b/c Lucene doesn't support boost values less than 0, but other than it is pretty straightforward using the TermVector support. At feedback time, get the TermVector for the top X documents and construct a new query using the frequencies of the terms for boosting (maybe multiplying by the alpha, beta, gamma parameters if you want). I seem to recall others posting that they have implemented similar things, so you may want to search the archive of this list.
I used the description in "Modern Information Retrieval" by Baeza-Yates and Ribeiro-Neto for the algorithm. >>> [EMAIL PROTECTED] 09/19/05 3:46 PM >>> Does anyone have experiences with relevance feedback and lucene or just knows some good websites? thx stefan --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
