This AI Plans Conference Schedules That Humans Actually Approve
Forget spreadsheets and headaches—AI can now build a whole event's agenda.
Organizing a conference's technical program is a giant, tedious puzzle. You have hundreds of research papers, and you need to group them into sessions that share similar topics, then fit those sessions into a schedule with hard limits on time and room availability. Human organizers spend days or weeks juggling these tasks, and even then, sessions can feel mismatched. TaxoConf, a new AI system from three computer scientists, promises to do this automatically—and do it so well that human organizers barely need to fix anything.
How does it work? The system first builds a map of topics based on the actual papers, like a family tree with branches for each research area. Every paper gets tagged with the tree's most specific topics that match. Then TaxoConf measures how close or far apart papers are in terms of those topics, and uses a mathematical process to assign papers to sessions in a way that minimizes the differences inside each session while obeying every practical restriction. The result: sessions that feel unified and programs that follow all the rules.
The numbers are impressive. When tested on data from four 2025 conferences, TaxoConf scored higher than previous methods on session coherence (4.62 out of 5), meaning papers in the same session were a good fit. It also matched human-curated sessions more closely than other automatic methods do. But the real proof came when TaxoConf was actually used to build the technical program for SIGIR 2026, a major computer science conference. Organizers accepted the initial AI-generated schedule almost as-is, keeping 91.8% of the oral presentation assignments and the entire poster program. They didn't need to fix a single rule violation. An on-site survey of 47 attendees rated overall program quality 4.26 out of 5.
What's the takeaway? This isn't just about academic conferences. It's a demonstration that AI can handle complex, constraint-heavy scheduling tasks that humans find tedious and error-prone. Think of wedding seating charts, corporate meeting tracks, or festival lineups—all involve similar puzzles. TaxoConf shows that with the right input, an AI can deliver an organized, satisfying structure that people accept and even enjoy. The system is publicly available, so others can adapt and use it.
- TaxoConf automatically groups research papers into coherent conference sessions, following strict rules about timing and limits.
- It was used to create the real program for SIGIR 2026, and organizers only made a few small edits—keeping nearly all oral presentation assignments.
- This points to a future where AI handles tedious event planning tasks, potentially saving weeks of human work.
Why It Matters
This means AI can now save event organizers enormous time while producing schedules people actually like.