The idea of "impact" and "impact-sorted posting list" should practically work with boolean model by approximation in the following way:
(1) Index Structure Inverted-Index : <term, posting-list>* posting-list: <impact, docID, occurrence*>+ (sorted by impact) occurrence: position (2) Retrieval Algorithm for boolean query "a AND b" set an impact threshold imp = imp0 get set of docs (each with impact) from a's posting list with impact of each doc larger than imp. get set of docs from b's ....... join the two sets by docID and aggregate the impacts of the same doc from the two sets. And sort the result set by impact. (for union query, union the two sets and aggregates impacts by docID....) if the result set size is too small, set a lower imp and **redo** the above. Note: If your major concern is relevance rather than recall for your retrieval (eg, on web), you will seldom hit the "redo" for conjunctive boolean queries. But often Lucene is used in an environment of small corpus and semi-structured data. Michael ____________________________________________________________________________________ Do you Yahoo!? Everyone is raving about the all-new Yahoo! Mail beta. http://new.mail.yahoo.com --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
