An ``expanded search'' is defined as a second or subsequent search for a given query, performed when the user selects the More Books button. Table 6a shows the number and percentage of queries involving expanded searches, and --- for those queries only --- compares the mean number of terms in initial and expanded searches.
As this table suggests, expanded searching normally takes place after the addition of some terms to the query via reformulation and / or expansion, but there are a few instances where users requested ``more books'' with an unchanged query. Conversely quite a lot of re-formulated or expanded queries are never used in an expanded search. Table 6b shows all the combinations of expanded searching ( ES) with query reformulation ( QR) and query expansion ( QE).
Table 6c attempts to quantify the benefits of expanded searching, i.e. whether it identified further relevant documents. It shows the number and percentage of queries where one or more relevant documents was selected by the user following an expanded search. It appears that expanded searching was of greater benefit to INSPEC users.
Table 6d goes a little further, and examines the benefits of expanded searching in the context of just those queries where either reformulation or automatic expansion (but not both) took place. The intersection sets are now quite small, particularly for INSPEC, but it does appear that automatic query expansion on its own was helpful for a group of LBS users.
Table 6e looks at the benefits of expanded searching in a slightly different way, by comparing the mean number of positive, negative and null relevance judgements, for initial and expanded searches. Once again, the figures refer only to queries involving expanded searches. For INSPEC users the mean number of positive relevance judgements went up slightly for expanded searches.
