Modeling Text Retrieval in Biomedicine

Given the amount of literature relevant to many of the areas of biomedicine, researchers are forced to use methods other than simply reading all the literature on a topic. Necessarily one must fall back on some kind of search engine. While the Google PageRank algorithm works well for finding popular web sites, it seems clear one must take a different approach in searching for information needed at the cutting edge of research. Information which is key to solving a particular problem may never have been looked at by many people in the past, yet it may be crucial to present progress. What has worked well to meet this need is to rank documents by their probable relevance to a piece of text describing the information need (a query). Here we will describe a general model for how this is done and how this model has been realized in both the vector and language modeling approaches to document retrieval. This approach is quite broad and applicable to much more than biomedicine. We will also present three example document retrieval systems that are designed to take advantage of specific information resources in biomedicine in an attempt to improve on the general model. Current challenges and future prospects are also discussed.