Deep learning based conference program organization system from determining articles in session to scheduling. Issue 6 (November 2022)

• Record Type: Journal Article
• Authors: Gündoğan, Esra; Kaya, Mehmet
• Is Part Of: Information processing & management. Volume 59:Issue 6(2022)
• Publication Date: 2022-11
• Subjects: Document similarity -- Clustering -- Scheduling -- BERT -- Organizing conference programs; Information storage and retrieval systems -- Periodicals; Information science -- Periodicals; Systèmes d'information -- Périodiques; Sciences de l'information -- Périodiques; Information science; Information storage and retrieval systems; Periodicals; 658.4038
• DOI: 10.1016/j.ipm.2022.103107
• Languages: English
• ISSNs: 0306-4573

Abstract

Highlights: An efficient conference program is automatically prepared for the participants and conference owners. A different clustering approach is proposed with as equal elements as possible in each cluster. An approach considering article content similarity in addition to word distributions is proposed for topic modeling. A method is proposed for the scheduling problem by assigning similar-themed sessions to different non-parallel timeslots. An improvement is achieved with the proposed method compared to the real conference programs. Abstract: It is very important to create the conference programs correctly in terms of timing and content by preventing problems such as being of articles that do not have a common topic with each other in the same sessions, the parallel of the sessions containing articles on the same topic. It greatly affects the efficiency of conference for participants. Currently, conference programs are organized manually. Considering the conference scope and the number of articles in that conference, it is a difficult and time-consuming process. In this study, an automatic solution to this problem is presented. The use of the SBERT method is provided a more accurate calculation of article similarities compared to baseline methods and is increased the success of other stages. Unlike classical clustering methods, an approach that clusters in such a way that there are equal numbers of data points in the clusters is proposed. In order to find the topic of the clusters determined as sessions, a topic determination approach is proposed that takes into account both keyword and article content similarities. Furthermore, with the proposed approach for session scheduling, the conference program has been planned more effectively by considering the parallel sessions. The ICTAI conference has been chosen to test the proposed approach. The proposed program is compared with both the real program and the programs created using Word2vec and Glove methods. With the proposed program, 10% improvement is achieved in terms of session similarity. In addition, parallel sessions are better planned with no conflicts compared to the real program.