Table 2a shows the total and mean number of terms in all initial searches. The short initial queries are highly typical of those seen in earlier investigations.
Table 2b shows the mean number of documents retrieved by all initial searches, and how they were distributed within ranked lists according to their degree of match. Documents are categorised according to whether they match the query ``well'', ``fairly well'', or ``not very well''. The figures in brackets indicate what percentage of the total retrieved list fell into each category. (Note that the distribution here is an artifact of the method used by the system to determine the boundaries --- see 2.3.2 for an account of the algorithms used in the current experiments.)
Clearly users were more successful at matching the system's language in searching the INSPEC database then in consulting the LBS library catalogue. A number of contributory factors may account for this difference including: the language of the subject domains (the precise terminology of computer science and IT vs the soft language of the social sciences), the subject-rich records of the INSPEC database compared to the brief catalogue records, and possibly the quality of the respective thesauri.