Continuous and Discrete Representations: A Reflection in the Context of IR
With the multitude of deep learning techniques being researched we might forget that information retrieval is about indexing, storage and retrieval of information. This talk is about representations of content, how they originated, and how they evolved over time up until today’s representations generated by deep learning. Representations influence how we index information and rank it in real-time in the most effective and efficient manner. From its early days IR research has combined continuous and discrete representations. It makes sense to reflect on their – sometimes conflicting - requirements. These include a rich and expressive model of content and its context, the normalization over different surface expressions, and the possibility of inferencing with or of aggregating content. In this respect interpreting continuous neural representations becomes an increasingly important task. The findings are illustrated with applications of ad-hoc retrieval, question answering and product search. We end the talk with promising directions for IR research.
Marie-Francine Moens is a full professor in the Department of Computer Science at KU Leuven. She leads a research team specialized in natural language processing and information retrieval. Her main direction of research is the development of novel methods for automated content recognition in text and multimedia using statistical and neural machine learning. She was the coordinator of the EU FP7 FET-OPEN project MUSE (Machine Understanding for interactive StorytElling, 2012-2015, EU FP7-296703) and is holder of the ERC Advanced Grant CALCULUS (Commonsense and Anticipation enriched Learning of Continuous representations sUpporting Language UnderStanding, 2018-2023, Horizon 2020-788506). She is currently associate editor of the journal IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) and was a member of the editorial board of the journal Foundations and Trends® in Information Retrieval from 2014 till 2018. In 2011 and 2012 she was appointed as chair of the European Chapter of the Association for Computational Linguistics (EACL).
Recommender Systems: Business Value and Measurements
Recommender Systems are one of the most visible and successful applications of Artificial Intelligence technology. Today, all major online sites provide personalized suggestions to their users. These recommendations can both be helpful for users and beneficial for the providers of these services. Academic research is mostly focused on the value of recommenders for the user; the value perspective of business, in contrast, is less explored. In this talk, we will review the various ways how recommender systems can create value, both for users and providers, and how this value contribution can be measured. We will furthermore discuss limitations of academic research in this context and outline potential ways forward.
Dietmar Jannach is a full professor of Information Systems at the University of Klagenfurt (AAU), Austria. Before joining AAU in 2017, he was a professor of Computer Science at TU Dortmund, Germany. In his research, he focuses on the application of intelligent system technology to practical problems and the development of methods for building knowledge-intensive software applications. In the last years, Dietmar Jannach worked on various practical aspects of recommender systems. He is the main author of the first text book on the topic published by Cambridge University Press in 2010 and was the co-founder of a tech startup that created an award-winning product for interactive advisory solutions.