Patterns of Document Access in Searching and Browsing

In the current work, we investigate the feasibility of using past experience to predict which documents will be accessed by users. Document access may be viewed as a surrogate measure of relevance, in which case the discussion here regards a method to improve retrieval effectiveness. But our main concern here is with users' access patterns per se. The prediction of future document accesses based on the past, is applied to two different IR services, (a) browsing and (b) keyword search. A straightforward method using conditional probabilities shows promise in both cases, while very different access patterns are observed for users of the two different IR services. These results have potential technical uses in improving document retrieval, and also shed light on the very significant differences between users of different IR-related services.