Research map / Large language models / AI and society
Promoting Research Collaboration with Open Data Driven Team Recommendation in Response to Call for Proposals
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Deploys ULTRA, a team-recommendation system that matches NSF/DST funding calls to researcher profiles using string, ACM-CCS taxonomy, and boosted relational bandit methods, scored by a goodness metric and evaluated via computational tests and a college-wide user survey.
Problem
Recommending research teams that respond to funding agency calls for proposals so that teams cover the demanded skills while balancing workload across candidate members, using skills latent in open data.
Approach
Formulates team recommendation for calls for proposals as a group recommendation problem, extracting skills from open RFP and researcher-profile data, matching via string, taxonomy (ACM-CCS), and boosted relational bandit methods, and ranking teams with a configurable goodness score combining redundancy, set size, coverage, and k-robustness.
Key results
- M3 boosted bandit reached average goodness 0.5295 at USC versus 0.0879 for random M0, while volume fell from 10 to 6 teams per researcher.
- In the 212-response user study, 90.09% of responses rated outputs relevant and 97.64% rated the tool useful.
- At IIT-R the same trend held: M3 averaged goodness 0.5835 with 1 team per researcher versus 0.0896 for M0 with 10 teams.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace random team formation with string matching (M1) | ULTRA · USC, 434 RFPs and 200 researchers | Average Quality: 0.0879±0.0290 -> 0.3673±0.1569 | Helped |
| Replace string matching with taxonomical matching (M2) | ULTRA · USC, 434 RFPs and 200 researchers | Average Quality: 0.3673±0.1569 -> 0.4097±0.1313 | Helped |
| Replace taxonomical matching with boosted bandit (M3) | ULTRA · USC, 434 RFPs and 200 researchers | Average Quality: 0.4097±0.1313 -> 0.5295±0.0816 | Helped |
| Replace random team formation with string matching (M1) | ULTRA · IIT-R, India; 100 RFPs and 46 researchers | Average Quality: 0.0896±0.0006 -> 0.4218±0.0011 | Helped |
| Replace string matching with taxonomical matching (M2) | ULTRA · IIT-R, India; 100 RFPs and 46 researchers | Average Quality: 0.4218±0.0011 -> 0.4292±0.0017 | Helped |
| Replace taxonomical matching with boosted bandit (M3) | ULTRA · IIT-R, India; 100 RFPs and 46 researchers | Average Quality: 0.4292±0.0017 -> 0.5835±0.0203 | Helped |
Surprising findings
- Users rated random M0 teams as very useful despite irrelevance, which the authors attribute to expanding new connections.
Limitations
- Human study was observational and unmonitored, so users did not necessarily try all use cases and methods.
- Dependence on open data that can be obsolete or inconsistent.
- No feedback on proposal success outcomes to validate recommendations.
Experimental setup
Datasets: NSF RFP archives USC faculty directories DST India RFPs IIT-R faculty profiles ACM CCS taxonomy
Benchmarks: USC RFPs and researchers (434 RFPs, 200 researchers) IIT-R RFPs and researchers (100 RFPs, 46 researchers)
Metrics: goodness score average team volume per researcher 5-point relevance Likert 5-point usefulness Likert
Baselines: M0 random team formation M1 string matching M2 taxonomical matching
Builds on: ULTRA relational boosted bandits ACM Computing Classification System
Closest papers by meaning
- CycleResearcher: Improving Automated Research via Automated Review (ICLR 2025 · similarity 0.83)
- ResearchTown: Simulator of Human Research Community (ICML 2025 · similarity 0.82)
- MLR-Bench: Evaluating AI Agents on Open-Ended Machine Learning Research (NeurIPS 2025 · similarity 0.82)
- Deep Research Arena: The First Exam of LLMs’ Research Abilities via Seminar-Grounded Tasks (AAAI 2026 · similarity 0.81)
- Position: AI/ML Influencers Have a Place in the Academic Process (ICML 2024 · similarity 0.81)
- AAAR-1.0: Assessing AI’s Potential to Assist Research (ICML 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.