Automatic Reviews' Assignments through Answer Set Programming

For the experiments, we selected the IRCDL dataset available from the 20th anniversary in which all the (315) publications of the first twenty editions were available. Authors, titles and keywords were extracted from the dataset. As per reviewers, we opted for the Program Committee members of the 20th edition of IRCDL.

The problem of automatically assigning reviews is a crucial task in conference management. Not onlyis it a time-consuming and challenging task, but it is also one of the most important peculiarities thatcontribute to the good/bad organisation of the conference, not to mention the degree of satisfactionof the people involved. During review assignments, many constraints come into play: the maximumnumber of papers per reviewer, the minimum number of reviews per paper, conflict handling and, lastbut not least, the similarity between papers’ topics and reviews’ interests. In this paper, we propose astrategy to map topics using the ACM Computing taxonomy, and a modelization of the problem in aConstraint Satisfaction setting relying on the Answer Set Programming logical framework. This strategy,although performing worse than top solutions in the state-of-the-art, showed the capability of managingmany real situations that have not been captured so far. Experiments demonstrated the goodness ofperformances for small/medium numbers of papers.