Improving Results for Focused and Relevance-in- Context Tasks

This is to certify that I have examined this copy of a master's thesis by Salil G. Bapat and have found that it is complete and satisfactory in all respects, and that any and all revisions required by the final examining committee have been made. Acknowledgements I would like to take this opportunity to thank several people who have contributed towards the successful completion of this thesis. I would like to thank Dr. Carolyn Crouch, for providing me an opportunity to work with her, her timely advice, and guidance throughout this thesis. I would like to thank Dr. Donald Crouch, for his suggestions and feedback during the important stages of my thesis work. I would also like to thank Darshan Paranjape and Sarika Mehta for being supportive co-workers as well as invaluable critics of my work. I would like to thank Aditya Mone, Nachiket Kamat and Vikram Malik for sharing their experience and helping me on various aspects of this thesis. Last but not the least, thanks to friends at UMD for the support and encouragement they provided during the two years of this degree. Thank you for all your blessing and wishes. ii Abstract Information retrieval focuses on organizing document-related data and retrieving useful information from it. Electronic documents on web can be structured, unstructured or semi-structured. Traditionally, information retrieval was concentrating on retrieving information from unstructured documents. But with the advent of the web and XML as a preferred standard for storing documents, the center of attention is now changing to retrieve information from structured and semi-structured documents. In such documents, the elements of the document at various levels of granularity can be retrieved instead of the document as a whole. This type of retrieval, for which elements of the document at various level of granularity are retrieved, is known as the flexible retrieval [1]. The main aim of this thesis is improved performance for the INEX 2007 Ad hoc Focused and Relevant-in-Context tasks. The Focused task requires systems to find the most specific (or focused) results that satisfy the information need. The Relevant-in-Context task requires systems to find the focused results that correspond to the relevant element in each relevant article. In this thesis, we discuss and evaluate various strategies to improve the performance of the Focused and Relevant-in-Context tasks while maintaining flexible retrieval as the backbone of the system.