An Article Similarity-Based Approach for Planning Conference Sessions.

Many conferences are held every year. The determination of the sessions for the accepted articles at these conferences is done manually by the organizers. As the number of articles increases, it becomes difficult to bring together related articles. In this study, an approach to the creation of conference sessions is presented. Sessions consist of articles on a common topic. To determine similar articles, semantic closeness between articles has been calculated using SBERT, SPECTER and SciBERT models. Clustering has been performed according to similarity scores, and the most similar articles have been grouped. Sessions are generally created with an equal number of articles. Therefore, unlike traditionally clustering methods, an approach that ensures that each cluster has equal data is proposed. When the session similarities of the real program and the proposed program are compared, an improvement of 9% in real program is achieved.