Table 5a shows the number and percentage of queries where the system used incremental query expansion to add new query terms, and the mean number of such terms added per query. (Note that by definition all these queries must have retrieved at least two relevant documents.) The expansion is greater in the case of INSPEC, probably because the longer documents provided more scope for automatic term extraction.
Tables 5b and 5c show ``snapshots'' of some selected queries at a particular point in a session, identifying terms which were typed in initially ( IQ), those added subsequently by the user ( US), and those added by the system via query expansion ( QE). Some thesaurus terms matched the initial query exactly and were used in the first search ( T1); some were explicitly selected by the user from the ``suggested words'' ( T2); however the majority were added automatically by the system through query expansion ( T3).
Note the tendency for query expansion to add proper names in LBS and semi-stop words in INSPEC. Users had the opportunity to delete unwanted terms, but may not always have noticed the addition of some lower-weighted terms once the list became fairly long.
(N.B. The list of terms shown in the first LBS example comes from a query for which nine independent searches were made. Other terms entered by the user (and subsequently deleted) were cargo, traders, and merchants.)
