Creating a conference schedule is a difficult task. Conference schedules consist of sessions, which contain papers that belong to the same field or subfield. Manually constructing such a schedule takes a lot of time, as each paper must be assigned to an appropriate subfield. This thesis presents a method for automating the schedule creation process. We use machine learning, natural language processing and network analysis to find papers with common research topics. Based on the similarities we group papers into predefined conference sessions using constrained clustering. We implemented the method as a part of a web application. To test the proposed method we created a database of academic papers from several machine learning conferences and labeled them manually with their research subfield. We tested each part of the method independently and obtained good results. The full method was tested on papers accepted to the ECML-PKKD 2017 conference. We obtained useful results that can be used as a starting point when creating a conference schedule.
|
