This project studies the benefits of combining two or more information retrieval schemes (data fusion). Methods are developed and tested using a large established test collection from the Text Retrieval conferences. The project is developing methods to predict when data fusion will be beneficial, and is developing methods that use non-linear rules of combination. The goal is to achieve a 15% relative improvement in the accuracy of predictions, and a 15% relative improvement in the effectiveness of the fused method. This project is motivated by the explosion of information in the World Wide Web, and includes development of specialized tests for comparing the new methods developed in this project to other "meta-search engines" already on the Web. For further information consult the Web site: www.scils.rutgers.edu/research/alexandria/SD.html
