Abstract
Building teams and promoting collaboration are two very common business activities. An example of these are seen in the TeamingForFunding problem, where research institutions and researchers are interested to identify collaborative opportunities when applying to funding agencies in response to latter's calls for proposals. We describe a novel deployed system to recommend teams using a variety of AI methods, such that (1) each team achieves the highest possible skill coverage that is demanded by the opportunity, and (2) the workload of distributing the opportunities is balanced amongst the candidate members. We address these questions by extracting skills latent in open data of proposal calls (demand) and researcher profiles (supply), normalizing them using taxonomies, and creating efficient algorithms that match demand to supply. We create teams to maximize goodness along a novel metric balancing short- and long-term objectives. We validate the success of our algorithms (1) quantitatively, by evaluating the recommended teams using a goodness score and find that more informed methods lead to recommendations of smaller number of teams but higher goodness, and (2) qualitatively, by conducting a large-scale user study at a college-wide level, and demonstrate that users overall found the tool very useful and relevant. Lastly, we evaluate our system in two diverse settings in US and India (of researchers and proposal calls) to establish generality of our approach, and deploy it at a major US university for routine use.
Introduction
Securing research funding increasingly requires assembling multidisciplinary teams to respond to complex Requests for Proposals (RFPs) from agencies like the NSF or NIH. However, forming the right team is often hindered by siloed information and communication gaps within large institutions. This paper introduces the "TeamingForFunding" problem, framing it as a group recommendation task. Unlike traditional single-item recommenders, our approach suggests complete, optimal teams that possess the specific skills demanded by an RFP while balancing institutional workload distribution.
The ULTRA System
We developed ULTRA (University-Lead Team Builder from RFPs and Analysis), an AI-driven system that leverages publicly available open data to bridge the gap between proposal demand and researcher supply. ULTRA evaluates candidate teams using a custom "Goodness Score" based on four weighted metrics: skill coverage, k-robustness, skill redundancy, and team size.
To generate recommendations, ULTRA employs four distinct algorithms:
- M0 (Random): A baseline grouping researchers randomly.
- M1 (String Matching): Exact keyword matching between RFPs and researcher profiles.
- M2 (Taxonomical Matching): Semantic matching using the ACM Computing Classification System (CCS) ontology to recognize related concepts.
- M3 (Boosted Bandit): A novel approach using relational contextual bandits and gradient-boosted regression trees to learn complex matching rules from relational data.
Experimental Evaluation
ULTRA was evaluated using two geographically diverse datasets: a US-based dataset from the University of South Carolina (434 RFPs, 200 researchers) and an India-based dataset from IIT-Roorkee (100 RFPs, 46 researchers).
Quantitative results revealed a significant "Quality vs. Quantity" trade-off. As algorithms moved from simple string matching (M1) to the sophisticated Boosted Bandit (M3), the system recommended fewer teams, but the average Goodness Score increased dramatically. For instance, in the USC dataset, M3 achieved a Goodness Score of 0.5295 (6 teams) compared to M0's 0.0879 (10 teams).
Results and User Study
A 28-day deployment at a major US university provided qualitative validation through 212 faculty evaluations. The results were overwhelmingly positive: 90% of respondents rated the recommendations as "very" or "somewhat" relevant, and 97.6% found the tool useful for identifying collaborative opportunities.
Interestingly, even the random baseline (M0) provided value by fostering serendipitous connections between researchers in different departments who were previously unaware of each other's work.
Conclusion
ULTRA demonstrates that AI can effectively automate the complex task of research team formation using open data. While the system's performance is currently tied to the quality of public faculty profiles, future work will focus on incorporating fairness constraints and human-in-the-loop refinements. The success of this deployment suggests that the underlying architecture could be scaled to other collaborative domains such as healthcare and emergency response.
Summary updated