Department of Computer Science & Engineering

Experience suggests that fully automated schema matching is infeasible, especially for n-to-m matches involving semantic functions. It is therefore advisable for a matching algorithm not only to do as much as possible automatically, but also to accurately identify the critical points where user input is maximally useful. Our matching algorithm combines several existing approaches with a new emphasis on using the context provided by the way elements are embedded in paths. A prototype tested on biological data (gene sequence, DNA, RNA, etc.)\ and on bibliographic data shows significant performance improvements from utilizing user feedback and context checks. In non-interactive mode on the purchase order schemas used in the COMA project, it compares favorably, and also correctly identifies critical points for user input.

Pre-2018 CSE ID: CS2004-0779