International distinction of ELMEPA’s research team in the RecSys IT competition

International distinction of ELMEPA’s research team in the RecSys IT competition
International distinction of ELMEPA’s research team in the RecSys IT competition

Recommender Systems are artificial intelligence systems that collect information about user preferences in order to provide users with predictions and suggestions for items unknown to them that are likely to be of interest to them. Such a suggestion system can be considered Google but also all the systems that make suggestions such as YouTube, Facebook, amazon, etc. and today show enormous commercial and research interest.

The global ACM Recommender Systems Conference (RecSys) focuses exclusively on recommender systems and has been held annually since 2007, being the most important scientific event in the field of Recommender Systems.

The competition is also being held annually for the tenth time in a row this year. In this year’s competition, the research teams were given real data from clothing purchases by the dressipi company, with the aim of the research teams predicting the final purchase of the user among approximately 5000 candidate products, taking into account the sequence of products visited in the online store as shown in Picture 1.

The full data set consists of 1.1 million online retail purchases sampled over an 18-month period, with the contest accepting results from 3/14 to 6/14. This year a total of 303 teams participated by sending results to the first phase of the competition and receiving information from the conference on their performance.

The second phase of the competition took place on a new data set where there was no information about the teams’ performance and from which the final ranking was derived, while the teams were asked to send the source code they created to be checked if it met the conditions and the limitations of the competition.

The ELMEPA team developed a novel hybrid method by combining nine probabilistic models with an LSTM neural network as shown in Figure 2. The used probabilistic models were trained to estimate transition probabilities and interactions between items, while the LSTM neural network learned the interactions between item features.

Work [1] submitted by the team was selected among the ten best to be published and presented at the competition with the possibility of remote presentation or in person at the conference that will take place in Seattle, USA from 18-23/9/2022. The distinction of the ELMEPA researchers is the only Greek participation that was distinguished this year, but also overall for the last 10 years of the competition, which proves the importance of the achievement for the Hellenic Mediterranean University in modern applications of data science.

The research team of the DataLab Laboratory of the Department of Administrative Science and Technology of ELMEPA that distinguished itself in the competition consists of professors Konstantinos Panagiotakis and Charalambos Papadakis. Konstantinos Panagiotakis is associate professor and chairman of the Department of Administrative Science and Technology

of Science (DET) of ELMEPA. He is also the director of the Department’s DataLab and a member of the Computer Vision and Robotics Laboratory of the ITE. Charalambos Papadakis is an assistant professor of the ELMEPA EMMY Department and director of the Telepi-

of societies and Information Technology of the Department. The success of the two ELMEPA professors in the field of recommender systems follows a long-term collaboration of more than 10 years with corresponding successes and more than 25 joint scientific publications, such as the publication made last year in the leading scientific journal of Artificial Intelligence according to the ranking of google scholar , Expert Systems with Applications scientific paper presenting a general method for improving the performance of recommender systems [2].

Also, during this period, the related scientific project Visit Planner is in progress, which supports their related research in recommendation systems. The purpose of the project is to offer the visitor the possibility of automatically composing a visit plan of the city of Agios Nikolaos, based on Artificial Intelligence technologies and the personal preferences of the visitor for a satisfactory experience in his available time and an initial version of the application is already available on the internet [3].

Bibliography

[1] C. Panagiotakis and H. Papadakis, Session-Based Recommendation by combining Probabilistic Models and LSTM, RecSysChallenge’22: Proceedings of the Recommender Systems Challenge, 2022.

Webpage: https://sites.google.com/site/costaspanagiotakis/research/recsys-challenge-2022

[2] C. Panagiotakis, H. Papadakis, A. Papagrigoriou, and P. Fragopoulou, Improving Recommender Systems via a Dual Training Error based Correction Approach, Expert Systems with Applications, 2021.

Webpage: https://sites.google.com/site/costaspanagiotakis/research/scor-dtec

[3]https://play.google.com/store/apps/details?id=com.netmechanics.vip

The article is in Greek

Tags: International distinction ELMEPAs research team RecSys competition

NEXT Journey through time – The weavers of Hermoupolis