Rastislav Papšo

Research areas: machine learning, deep learning, recommender systems, computer vision

Position: PhD Student 09/2022-12/2023

Rastislav is a research assistant focusing on the topics of machine learning, deep learning, recommender systems, and computer vision. He is part of the Web & User Data Processing team.

He holds master’s degree in Computer Science from Faculty of Management Science and Informatics, University of Žilina. He graduated with distinction. During his studies, he participated on several research projects related to application of deep learning algorithms in computer vision. Both his bachelor’s (Face identification based on a small amount of data) and master’s (Size reduction of deep neural networks in classification tasks) theses were related to these topics as well.

In addition, he has more than two years of practical experience in software engineering and data science, that he obtained during his internships and part-time jobs he attended during his studies.

PhD topic: Recommender and adaptive web-based systems

Supervising team: Michal Kompan (KInIT), Luigi’s Box (industry partner)

The recommender systems are an integral part of almost every modern Web application. Personalized, or at least adaptive, services have become a standard that is expected by the users of e-commerce. In recent years, self-supervised learning and neural networks have become state-of-the-art. We focus on the improvement of the performance of session-based recommenders when users have no or little previous activity. To improve the precision of the recommender we consider additional user and item features, as well as the sequence itself. 

There are several challenges which have to be addressed. In e-commerce there are often additional business requirements such as the diversity or availability of the recommended items. Moreover, the item’s description is often missing or includes only basic information. Similarly, in the recommender as a service setting, the performance costs play an important role. Together with no user history makes the recommendation challenging in many ways.