Scheduling the technical sessions of scientific events is an arduous task commonly faced by many organizers worldwide. Due the particularities of each conference, there is no consensus regarding the problem definition, and researchers have tackled each specific case individually. Despite their distinct characteristics, one often expects the sessions to be composed of presentations of similar scope. This natural assumption led us to define a basic yet sufficiently general version of the problem that aims at maximizing the benefit of clustering papers with common topics in the same session, while leaving the particularities of the event to be addressed by means of side constraints. Three mathematical formulations based on integer linear programming are proposed for the problem, which in turn is shown to be NP-hard. The first model consists of a compact formulation, whereas the second and third models serve as underlying formulations for branch-and-cut (BC) and branch-cut-and-price (BCP) algorithms, respectively. Computational experiments were carried out on real-life and artificial instances derived from two conferences, considering several scenarios. While the compact formulation and the BC procedure could solve instances with moderate size, the BCP approach managed to produce tight bounds for instances with up to 163 presentations.
Bulhões, T., Correia, R., Subramanian, A. Conference scheduling: a clustering-based approach. European Journal of Operational Research, Forthcoming. https://doi.org/10.1016/j.ejor.2021.04.042