I’ve been involved in the organisation of the PRASA-Robmech conference a few times, and it’s usually a painful experience. In an attempt to simplify things, we dashed together an automatic conference scheduler, which also gave me a chance to explore latent topic models. The scheduler takes a conference schedule skeleton and full text papers, then uses latent Dirichlet allocation to extract paper feature vectors. The algorithm we developed then attempts to allocate the papers to sessions to minimise within session distance between paper features, while maximising parallel session distance, so as to avoid potential talk conflicts.
This works really well, although it turns out manual intervention is often required to accommodate special requests, and to avoid the inevitable catch-all conference track. We used this algorithm to schedule the 2016 PRASA-RobMech Conference when I acted as programme chair, but tested our approach on 2014 and 2015 papers.
The code is available here.
Burke, Michael, and Deon Sabatta. “Topic models for conference session assignment: Organising PRASA 2014(5).” Pattern Recognition Association of South Africa and Robotics and Mechatronics International Conference (PRASA-RobMech), 2015. IEEE, 2015.