ECIR 2017 conference review

Aberdeen – granite, oil, gas, and home of The Grill

For its 39th edition, the European Conference in Information Retrieval (ECIR) returned, after 20 years, to Aberdeen, Scotland. The conference was organized by the School of Computing Science and Digital Media at Robert Gordon University (RGU) and took place in two separate venues: tutorials and workshops were held at RGU’s Garthdee campus, while the main conference, as well as the industry day, were hosted by the Aberdeen Exhibition Conference Center. The wonderful backdrop of ECIR was the city of Aberdeen, also known as the Granite City; the origin of this nickname can be traced to the high number of buildings in the city that incorporate grey granite.The granite sparkles under the sunlight; thanks to five days of almost perfect weather, conference attendees were able to enjoy this beautiful effect.

ECIR 2017 received 135 full paper, 101 short paper and 12 demonstration submissions. Of those, 27%, 35%, and 58% of them were accepted for publication (36 long papers, 35 short papers, and 7 demos). Compared with previous years’, the format of the conference was significantly altered: instead of three days of parallel tracks, the organizers opted for a single track program. To accommodate this change, full paper presentations were partitioned in two groups: 6 papers were allotted 30 minutes, while the remaining 30 were allotted 10 minutes. All short papers authors were also given an opportunity to briefly introduce their work in a 2 minutes booster presentation session. Finally, for the first year, ECIR included a doctoral consortium.

The radical change in the structure of the conference turned out to be a successful one: the number of people attending the presentations was high, and conference participants were very engaged with works presented in all sessions.

Day 1: Sunday, April 9th

Similarly to previous years, the first day of ECIR was dedicated to workshop and tutorials; in total, two full-day workshops, two half-day workshops and one half-day tutorial took place.

Day 2: Monday, April 10th

After a warm welcome from program chairs Joemon Jose and Claudia Hauff, Laura Dietz from University of New Hampshire delivered the first-day keynote. Her presentation focused on methods for improving search results with structured data to answer complex queries. In particular, after a brief introduction on entity linking, Laura presented successful approaches to search results augmentation, as well as the challenges in matching the unstructured text in search results with structured information from knowledge graphs. How to deal with missing or mistaken entities? What are the dangers of mining the knowledge graphs for related entities? Laura concluded her keynote by mentioning some still open problems in the field, encouraging other researchers to pursue them.
The deck for her keynote is available on her homepage.

Two full paper presentation sessions followed the morning keynote; one took place before the lunch break, while the other took place after. Because presenters in the pre-lunch session were allotted a 10 minutes slot to explain their work, conference organizers encouraged authors to present posters summarizing their work during the launch session. This approach turned out to be very successful, as many conference attendees visited the poster booths while enjoying their lunch and engaged in thoughtful discussions (presenters might disagree, as they were only able to gaze lunch from afar!)

The first day of the conference concluded with the short paper poster session. As mentioned above, short paper authors were also given the opportunity to introduce their work to all conference attendees, albeit for just two minutes, in a paper booster session. Despite the strict time constrains, all presenters did a fantastic job keeping their presentation concise and insightful, giving the audience an opportunity to get a complete overview of the numerous posters at the conference.

Day 3: Tuesday, April 11th

Jaime Teevan – KSJ Award winner and great speaker

The award ceremony for the 2016 Microsoft BCS/BCS IRSG Karen Spärck Jones Award was at the top of the agenda for the second day of the main conference. This year, the award was given to Jaime Teevan, Principal Researcher at Microsoft Research and an Affiliate Associate Professor at the University of Washington. The award panel decided to recognize Jamie for her “technically strong and exceptionally creative contributions to the intersection of information retrieval, user experience and social media”.
This was the last award given by the committee chaired by Stefan Rüger; starting next year, Udo Kruschwitz will lead the nominations for the 2017 award.

Following the presentation of the KSJ Award, Jamie presented her work in recurring queries in online search. Her work stems from the observation that online searchers often repeat the same searches in order to re-retrieve content they have seen before. This presents both of a problem for search engines, as their are often designed to promote novelty, as well as an opportunity for improving the search experience of its users. Her presentation spanned 20 years on research on web search engines, from the early days of Infoseek to the present. The take away statistic from Jamie’s keynote? 43% of all queries and clicks are repeated, so addressing repeated search is a must for modern search engines.

Panel time!

After two long poster presentation sessions, the scientific program for the day ended with what turned out to be a very lively panel about recent artificial intelligence advances and how they relate to information retrieval. Aptly named “IR & AI”, the panel featured Laura Dietz, Alex Hauptmann, and Julio Gonzalo; it was moderated by Maarten De Rijke. The focus of the panel quickly settled on discussing what is the impact of advances in a field that values methods (AI) would be on a field that is very focus on the processes (IR). While some believe that the rigor in conducting experiments and richness of topics studied in information retrieval would keep this field alive despite the rapid change in approaches, others are concerned by the fact that systems that require big data might sweep away the nuanced topics many IR researchers care about. In particular, the fear is that models trained on large amount of data and designed to solve common problems would shadow the efforts of many researchers who try to improve less frequent yet important use case scenarios. Finally, the panel briefly touched the topic of whether significant changes are needed in the scope of IR conferences to accommodate the growth of IR-related research that leverages recent advances in deep learning. Coincidentally, notifications for SIGIR 2017 started coming in during the panel, showing a sharp increase in number of accepted papers that leverage neural approaches.

The social dinner was held at Rox Hotel, where an excellent dinner was accompanied by traditional Scottish music. During the dinner, a number of awards have been given. The best paper award was given to “The Effects of Search Task Determinability on Search Behavior” by Rob Capra et al. from University of North Carolina at Chapel Hill. The best student paper award went to Anje Fang et al. from the Terrier Team at University of Glasgow for their work

Ceilidh – what else? It’s Scotland!

on exploring bayesian inference-based LDA for social media datasets.
Finally, “Similarity Measures for Short Segments of Text” by Donald Metzler et al., originally published in ECIR 2007, won the test of time award. For the few who sticked until the end, the social dinner closed with an intense session of Scottish social dancing that left everyone pretty much breathless.

Day 4: Wednesday, April 12th

The last day of the main conference opened with a keynote by Alexander Hauptmann from the School of Computer Science at Carnegie Mellon University. Alex discussed the tradeoff between accuracy and performance in video search systems. State of the art retrieval systems have impressive performance, but often rely on expensive techniques that are too computationally expensive to be used at scale.

At the end of the two long paper sessions scheduled for the day, ECIR 2017 came to a close. After a few words from the general chairs, Georges Quenot took the stage to give a preview of next year’s ECIR, which will be help in Grenoble, France. All readers interested in submitting or attending should visit the conference website.

About Luca Soldaini
Luca Soldaini

Luca Soldaini is a Ph.D. candidate at the Department of Computer Science at Georgetown University and a member of the Information Retrieval Lab. He earned his undergraduate degree in Computer Engineering at University of Florence. Luca's research interests include text mining, information retrieval, and NLP applications for the medical domain. In particular, he is focused on improving access to medical information by bridging the knowledge gap between health information seekers and medical content.

Leave a Reply