An Exploration of Retrieval-Enhancing Methods for Integrated Search in a Digital Library

Integrated search is defined as searching across different document types and representations simultaneously, with the goal of presenting the user with a single ranked result list containing the optimal mix of document types. In this paper, we compare various approaches to integrating three different types of documents (bibliographic records for articles and books as well as full-text articles) using the iSearch collection: combining all document types in a single index, weighting the different document types using priors, and using collection fusion techniques to merge the retrieval results on three separate indexes corresponding to each of the document types. We find that a properly optimized retrieval model on a single combined index containing all documents without any special treatment performs no worse than our weighting and fusion methods, suggesting that more work is needed on alternative approaches to integrated search.