Handling Multi-faceted Queries

Some queries involve a combination of three or more topics or facets. In a boolean query language, separate topics can be explicitly linked using the AND operator; in a probabilistic system it is necessary to ensure that each is given its due weight regardless of variations in term frequency. Following thesaurus navigation, the form of the enhanced query should also retain faceting information.

The final version of the thesaurus navigation software allowed individual topics to be entered and explored independently. The list of terms submitted to the Okapi retrieval system was grouped accordingly. Query enhancement involved combining all terms from the same topic using boolean OR, finding weights for each topic group, then applying the normal "best-match" algorithm across the groups.

The following examples show how some users chose to partition their queries:

During thesaurus navigation, users with faceted queries collected more terms than those without - on average 15 terms as against 9. This is perhaps not surprising in view of the fact that they had more potential paths to follow.

As in previous experiments, the performance of the original query was compared with that of the enhanced query - this time the faceted form was included in the comparison where applicable.

The intention was to test:

As a result of these experiments, future work will attempt to divorce query faceting information from the probabilistic search process, and to use it only in the selection and display of documents retrieved by a probabilistic search. For a preliminary discussion see the details of the Enquire project.