Experiments with the TREC data have also shown that much of the evidence
that is crucial for effective navigational search is not important for topical
searches. In fact, the only features needed for topical search are the simple
terms and proximity terms for the body part of the document. The other features
do not improve effectiveness, but they also do not reduce it. Another difference
between topical and navigational searches is that query expansion using
pseudo-relevance feedback was found to help topical searches, but made navigational
searches worse. Navigational searches are looking for a specific page, so it is
not surprising that smoothing the query by adding a number of extra terms may
increase the “noise” in the results. If a search was known to be in the topical category,
query expansion could be used, but this is difficult to determine reliably,
and since the potential effectiveness benefits of expansion are variable and somewhat
unpredictable, this technique is generally not used. Given that the evidence
needed to identify good sites for transaction searches seems to be similar to that
needed for navigational searches, this means that the same ranking algorithm can
be used for the different categories of web search.