Keyword search is the accepted paradigm for data retrieval and exploration on the web. Dynamic, data-driven websites serve content from data repositories, which provide only limited support for keyword search. Social networks and other private data repositories cannot be indexed by ordinary web search engines, yet users desire the same, high-quality search results they have come to expect from Internet searches. Despite a decade of academic research, relational keyword search strategies for data graphs have not yet been shown to have either the scalability or the usefulness of the ad hoc search techniques widely used on today's websites.

This project focuses on the evaluation of relational keyword search systems. Our work challenges the ad hoc evaluations reported in the literature, evaluations that lag far behind the best practices developed in information retrieval (IR) for evaluating retrieval systems. Standardized evaluation will enable the same rapid progress in this field that the IR community experienced following the inception of the Text REtrieval Conference (TREC). Our work requires the construction of real-world workloads to benchmark the efficiency and effectiveness of existing systems. The evaluation framework will then guide the development of novel techniques for ranking results and query processing.

Successful e-commerce depends upon many elements, not the least of which is well-trained computer scientists to design, maintain, repair, and innovate e-commerce technology. Experience points in the direction of a significant need for expanded education focused on the security of e-commerce systems. This need is consistent with a broader trend in computing: security has been identified as a grand challenge in the field and the weakest link of many of the nation's most critical applications and services. This project focuses on developing teaching materials and techniques specifically related to security in e-commrce.

A problem endemic to a diverse range of organizations (including medicine, engineering, and government intelligence) is knowing who to collaborate with on a given problem. Crosspoint is a web-based collaboration service designed to enhance interaction among problem holders. Users submit requests for information through Crosspoint's web interface. The search engine quickly identifies subject matter experts from the profiles created by each user. Electronic feedback, organization, and past experience quickly narrow the list of potential collaborators. Following collaboration, the experts provide the desired information to the original problem holder.

Embedded systems, especially those in safety-critical or hard real-time environments, typically require that timing constraints be met. To guarantee these systems will meet deadlines, the worst-case execution time (WCET) of each program must be known. The process of determining the WCET of a program is known as timing analysis. Knowledge of the WCET can be used to dynamically scale voltage when a scheduler detects future slack time. This facilitates power savings, an especially important aspect in embedded systems. Image processing provides another useful application for parametric timing analysis, since image dimensions may not be known a priori.

Static timing analysis has traditionally required loops to contain a constant number of iterations so analyzers may produce constant worst case execution bounds. Such constraints on input programs make this form of timing analysis numerical: the number of loop iterations is constant as is the final result from the timing analyzer. Parametric timing analysis allows the number of loop iterations to be unknown at compilation as long as this value may be written as an expression. Such flexibility expands the class of programs which may be analyzed. Instead of providing a constant upper bound for a loop, a symbolic formula is created using the expression representing the number of loop iterations. The formula may be evaluated later to obtain the execution time for any given input.