"Precision Content Retrieval": Ambroziak and Woods (1998)

• The goal here is to improve information retrieval by term expansion in a taxonomy and by returning just the relevant passage of a document, rather than a whole document.
  • Starting hypothesis: multiword phrases can be useful if we "do more with them than simply look for exact matches".
  • The "ConceptStore" program starts with a WordNet-like taxonomy, and extracts phrases from documents to build a "phrasal taxonomy" on the fly.
  • Queries are matched to phrases near them in this phrasal taxonomy. Some rearrangements of words in a query are permitted when matching a phrase, but the phrase then incurs a penalty score (better matches have smaller penalty scores).
  • The user can view the taxonomy, and revise the query, making it more general or more specific if appropriate.
• Results are claimed to be much better (on a task involving queries about the UNIX operating system) using these methods, as opposed to garden-variety IR tools. But they point out that improvement is not "monotonic"; every method gets some results that others miss.

(more on Precision Content Retrieval)

(back to general IR slides)

(on to collocations)

(return to syllabus)