Query Expansion
Query expansion involves taking documents judged relevant by the user as
the basis for a new search to find others which are "similar". Words and
phrases are extracted from the relevant documents, weighted, and added to
the existing search terms to generate an expanded query. Early evaluation
of this facility (see Okapi '88) showed that it
was effective for finding documents on the same topics as those already
chosen.
Recent Okapi studies have examined the way in which users cope with
this facility, and how the interface should be designed to present the
concept of query expansion. It can operate in two different ways:
- In Automatic Query Expansion (AQE), as described
above, Okapi extracts words and phrases from relevant documents and
presents the user with a new list of document titles together with the
words and phrases that were used to find them.
- In Interactive Query Expansion (IQE), Okapi
extracts words and phrases from relevant documents and presents them to
the user who decides which of them to include in a new search. The choice
can be made by:
- selecting only those terms which are required, or,
- deleting those terms which are not required.
The methods were compared in the Okapi '94
experiments. Results suggested that giving users more control over the
inclusion of new words and phrases in an expanded search was not always
beneficial. Few subjects took up IQE, and those who did tended to be
highly selective, making the expansion facility less effective.
A later research project, Enquire Okapi,
re-examined query expansion methods, and investigated how best to
combine AQE and IQE with thesaurus searching, so as to gain maximum
benefit from user relevance judgements in improving search performance.
The new interface introduced the concept of dynamic or
incremental query expansion.