How successful has modeling been? As an example, ranking algorithms for
general search improved in effectiveness by over 100% in the 1990s, as measured
using the TREC test collections. These changes in effectiveness corresponded to
improvements in the associated retrieval models. Web search effectiveness has
also improved substantially over the past 10 years. In experiments with TREC
web collections, the most effective ranking algorithms come from well-defined
retrieval models. In the case of commercial web search engines, it is less clear what
the retrieval models are, but there is no doubt that the ranking algorithms rely on
solid mathematical foundations.