A recent (May 2010) excellent doctoral dissertation was written by Lewis Church here at UNC-CH, Combinatoric Models of Information Retrieval Ranking Methods and Performance Measures for Weakly-Ordered Document Collections, addresses analytic models of retrieval.

List of publications on Retrieval by Robert Losee
See also the list of publications on Natural Language Processing (many of which address retrieval concerns) and publications on Classification and Organizing Information for Browsing .

The past and future: My recent work has developed analytic (e.g. non-experimental) methods for predicting retrieval performance. I have emphasized the simple case where there is only a single term in the query. This has enabled me to make statements (e.g. when does incorporating relevance feedback or including part-of-speech tags improve IR performance) that are independent of individual test databases. Present and future work is or will consider distributed data, different ranking and learning algorithms, and more elaborate multiple term queries.

Students who have completed a semester or two of master's level work and have taken the related introductory courses in this area might want to consider taking an independent readings course if they believe that this area represents their career focus.