People of ACM - Dietmar Jannach

March 22, 2022

What led you to your current work of developing machine learning algorithms for recommender systems?
Our current algorithmic work is on what is called session-based recommendation. Here, the challenge is to develop systems that are able to adjust the recommendations according to the most recent observed user actions, e.g., a purchase on an e-commerce site, in real-time. This problem is highly relevant in practice, but has only recently attracted more research interest. Our own research in this area was inspired by a cooperation with an industry partner, who faced the challenge of correctly identifying the consumer's short-term needs from only a few observed and often noisy interactions

Your 2019 paper “Are we really making much progress? A worrying analysis of recent neural recommendation approaches,” (co-authored with Maurizio Ferrari Dacrema and Paolo Cremonesi) won the RecSys Best Paper Award and is one of your most-cited works. The paper raises questions about reproducibility and about current research practices to demonstrate progress in the area of machine learning algorithms in recommender systems. How might the recommender systems field change course in this area?
Our research in this area continues a number of earlier works by others who were worried about the level of reproducibility of published research findings, both in the area of recommender systems and in applied machine learning in general. There are various reasons for the observed phenomena, which may partially be tied to underlying issues of our publication culture and academic incentivization system. In the short term, we believe that increasing the level of reproducibility, e.g., by establishing a culture where sharing all artifacts that were used in experiments is normal, may help to avoid many similar issues in the future.

What has surprised you the most about how the recommender systems technology has developed since you began working in this field?
In 2009, a large machine learning competition organized by Netflix ended. By that time, one could have thought that the recommendation problem is almost solved now that we have sophisticated algorithms that can very precisely predict what consumers will like. It however turns out that still many questions need much more research. This is the case for example in the area of human-computer interaction or in terms of understanding how such systems may positively or even negatively affect consumers and providers. Moreover, there is a lot of promise in modern machine learning algorithms, for example, when it comes to building more interactive and conversational forms of recommender systems.

You announced the launch of TORS at the RecSys 2021 conference. Why is a dedicated journal for recommender systems research needed? What are your goals for TORS?
Building an effective recommender system in practice is a highly challenging and multi-faceted problem. From a computer science perspective, for example, we need algorithms that are effective for a given problem setting and we need appropriate user interfaces that allow users to conveniently explore the space of options and that help them to make better decisions. However, to be effective, these technical solutions must be informed by research in other fields such as psychology, consumer behavior, or marketing. With this new journal, we created a publication outlet that reflects the multi-faceted nature of the problem and should serve as the premier venue where people from academia and industry will find the latest research on recommender systems.

Dietmar Jannach is a Professor at the University of Klagenfurt in Austria. He has authored more than 150 publications in areas including recommender systems technology, knowledge-based systems development, constraint-based systems, semantic web applications and web mining, and software engineering. Among his publications, Jannach is a co-author of the book Recommender Systems: An Introduction. His current line of research is focused on the design and evaluation of machine learning algorithms for recommender systems and on the impact and value of recommender systems in practice.

His awards include the Advancement Award by the Province of Carinthia (Austria) in the area of Technical Sciences, and several Best Paper Awards, including the ACM Best Full Paper Award at ACM RecSys 2019. He is the Co-Editor-in-Chief (along with Li Chen of Hong Kong Baptist University) of the new journal ACM Transactions on Recommender Systems (TORS). TORS is the first journal of its kind dedicated exclusively to different aspects of recommender systems. The editors are now accepting submissions and the first issue will be published in 2022.