Impact of Terminal Node Processing on Element Retrieval

This is to certify that I have examined this copy of master's thesis by Vikram Malik 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 First and foremost, I would like to thank my advisor Dr Carolyn Crouch, for her support and guidance for the last two years. She was patient and always willing to share her immense experience in the field. I would also like to thank Dr Donald Crouch and Steve Holtz for their valuable suggestions and feedback time and again. I would like to thank Jim Luttinen for promptly taking care of the system failures we had this year, even while he was on vacation. I would also like to thank Lori Lucia and Linda Meek for all their administrative help and words of encouragement. Finally, I would like to thank the Department of Computer Science and my friends here in Duluth for making the last two years an unforgettable experience for me.Abstract Information retrieval strategies of earlier years were developed around the idea of retrieving entire documents in response to a user's request. But with widespread use of markup languages like XML (Extensible Markup Language) for representing documents, the idea of retrieving at the element level evolved. And with the continuous growth in XML information repositories available today, the focus on developing effective element retrieval strategies has never been stronger. Our element retrieval strategy, called flexible retrieval [7], requires maintaining only a single index of the leaf nodes of documents and is designed to work with a strictly structured collection of well-defined and non-overlapping element nodes. In this research, we examine the feasibility of extending this technique to the real-world, semi-structured Wikipedia collection [4]. We tune for slope and pivot values for this collection against the new set of cumulative gain based measures [6] adopted by INEX in 2007. Experiments comparing the results obtained from our technique with retrieval done against an overlapping index of all document elements (the all-element index) are detailed. We describe experiments to determine the impact of processing terminal nodes prior to evaluation and show that it dramatically improves evaluation results in most cases.-iii