In its 12th edition, the European Summer School in Information Retrieval (ESSIR) returned to its Italian roots in beautiful Milan, where it took place at the University of Milano-Bicocca between July 15th to July 19th, 2019. Maristella Agosti (University of Padua), who was organizer of the first summer school back in 1990, again co-organized the event, together with Nicola Ferro (University of Padua) and Gabriella Pasi (University of Milan-Bicocca).
Over 40 students from as far as Chile and China attended the event, which is primarily geared towards early-stage researchers. During the week, attendees could enjoy talks on a wide variety of topics held by established members of the IR community. For many attendees, especially more senior ones, the schedule of the summer school also fit neatly with further travel plans to attend SIGIR 2019, which started on the subsequent Sunday in Paris. Co-located with ESSIR was the 9th PhD Symposium on Future Directions in Information Access (FDIA), this year with PCs Ingo Frommholz (University of Bedfordshire), Haiming Liu (University of Bedfordshire) and Yashar Moshfeghi (University of Strathclyde). In total the event received 25 submissions, out of which 7 were chosen for oral presentation, and 12 displayed during a poster session.
First Day
After a warm welcome by the organizers, Maarten de Rijke (University of Amsterdam) opened the summer school with his talk, introducing the students to a theoretical framework of IR research that allows the categorization of later talks. Especially his quote “Information Retrieval [is] getting the right information to the right people in the right way” was highlighted quite a few times throughout to week to illustrate the relevance of the users in the equation. A more historical overview was then also presented by Bruce Croft (University of Massachusetts Amherst), who commented on the past and present research opportunities in Information Retrieval.
During the afternoon sessions, Nicola Ferro (University of Padua) and Julio Gonzalo (UNED) presented the difficulty of proper evaluation in IR. Both of them also touched on an issue that was strongly echoed by subsequent speakers: Reproducibility is one of key challenges – not only in IR, but in general – of our time and needs to be addressed properly, to ensure the meaningful advancement in research. The concluding icebreaker opening reception provided the perfect way for students to meet each other and the lecturers in a less formal setting.
Second Day
During the morning session, the topic was quite different from the previous day and focusing on the user-oriented aspect in IR. While Nick Belkin (Rutgers University) introduced various perception models for the Information Retrieval process, starting from the first important works by Cleverdon, Diane Kelly (University of Tennessee) walked the students through a history of different user interfaces, all the way to modern smartphone interfaces.
During the afternoon session by Claudia Hauff (TU Delft), students learned the value of a well-tuned baseline, and how neural IR models are not the solution to every problem, especially with respect to explainability and reproducibility. Gabriella Pasi (University of Milano-Bicocca) concluded the day by introducing the notion of context in IR, as to avoid a “one size fits all” approach.
Third Day
After two intense days of more theoretically inclined lectures, Wednesday brought two refreshing morning sessions by Nicola Tonellotto (ISTI, CNR) and Carlos Castillo (Universitat Pompeu Fabra). Again, the discrepancy between the research in industry and academia became a central aspect of the talk by Nicola Tonellotto, who offered a glimpse into the issues of production systems at scale, and what measures can be taken to implement many of the great ideas and techniques in practice. Carlos Castillo on the other hand talked about mining social networks, and also warned the students to not blindly follow correlations or data that could be obtained way easier.
During the afternoon sessions, the first part of the FDIA symposium took place. Eight papers were selected for oral presentation, covering a wide range of topics including authorship verification (Silvia Corbara), a Bayesian Personalisation Ranking (BPR) based recommendation system (Siwei Liu) and exciting research into cortical activity of relevance (Zuzana Pinkosova). The award for Best Paper & Presentation was later given to Dennis Dosso (University of Padua) for his work on ‘A Keyword Search & Citation System for RDF Graphs’.
Fourth Day
Henning Müller kicked off proceedings on Day 4 of the summer school, focussing on a broad spectrum of research directions under the umbrella of Medical (text and image) IR. Henning presented a compelling case for the need to improve how information is presented to medical staff, including some novel information fusion techniques, combining both image and textual features. Paolo Cremonesi (Politecnico di Milano) followed up with a fundamental lecture on a very related topic – recommender systems. During his session, he made sure to cover a wide variety of problem variations in the field and formalize the problem to the students.
The afternoon sessions promised to be rewarding for students and did not disappoint. Firstly, a panel including Ricardo Baeza-Yates, Bruce Croft, Diane Kelly & Henning Müller (chaired by Maristella Agosti) shared their wealth of knowledge and experience with the students. On the other hand, students also had the chance to address some of their suggestions for the improvement of future ESSIR editions. Then attention shifted back to the students for the FDIA poster session. This year, twelve posters were presented with Emma Gerritse (Radboud University) picking up the Best Poster award for her work on debiasing word embeddings.
Continuing the conversations from the previous sessions at a social dinner in a typical Italian restaurant, further exchange was facilitated by the organizers and emphasized the almost family-like atmosphere at the event. In addition, the FDIA chairs formally awarded the Best Presentation/Post award to the respective recipients, although they stressed how hard the decision was due to the number of great submissions.
Last Day
During the last day, Emine Yilmaz (University College London) talked about understanding and inferring user needs, for which she explained the difficulty of multitasking from a system design point of view. Especially tracking user behavior across multiple sessions is still a topic worth exploring, and sees further research put out each year. As the final talk, Ricardo Baeza-Yates’ (NTENT/Northeastern Univeristy) session on biases in web search and recommenders was already hotly anticipated by many attendees, as the topic had already been mentioned by many of the other speakers, and seemed to fit very well in the context of current research interests. He convincingly laid out the proposal that no one is immune to all biases, and that the sheer variety of problems – broadly categorized in cultural, statistical or cognitive biases – needs a very determined researcher to realize the interplay of different factors in datasets.
To conclude a week of intense lectures, this years’ edition introduced a small student challenge for the participants, during which they were asked to present a potential research topic in small groups. Despite the lack of time to prepare an in-depth proposal, many of the groups already had a good idea of what they could do, especially focusing on the aspects and hot topics they learned about during the school’s sessions. The time to depart came all too soon after, and it seems everyone left exhausted, but also very excited to continue their PhD journeys.
Impressions
As a PhD student that just recently started, ESSIR & FDIA were the perfect place to be in order to get a broad foundation in IR, especially since many universities still do not have dedicated (or mandatory) IR lectures in place. Every single teacher was extremely motivated, and really made a lasting impression on my with their enthusiasm for the field. Aside from the obvious learning experience, meeting peers that are in a somewhat similar position helps immensely to start building a network of potential future collaborators and friends. Especially facilitated through the symposium’s presentations, it became clear what individuals are working on or planning for their future experiments.
Thanks to the engaged committees for both ESSIR and FDIA and with the help from a number of sponsors, it will be certainly an unforgettable experience.