The 40th edition of European Conference on Information Retrieval, ECIR 2018, was held in the picturesque city of Grenoble, France. The conference was hosted in Minatec, which boasts of a nice view of the Alps right from the conference venue. A perfect weather and central location meant the attendees could indulge in some sightseeing in the post-conference hours. Éric Gaussier (University Grenoble Alpes), Lorraine Goeuriot (University Grenoble Alpes) and Georges Quénot (CNRS) were the general chairs for ECIR 2018.
ECIR 2018 received 171 full papers and 112 short papers, out of which 39 full papers (23%) and 29 short papers (35%) were accepted. Key highlights of this year’s ECIR were a special focus on industry with all the three keynote speakers being non-academicians, an equally unprecedented amount of interest shown by the attendees towards the industry sessions (more on that later), reproducibility studies getting a significant boost in form of the best paper award and ECIR receiving A ranking in the prestigious CORE ranking. A major announcement about a collaboration between BCS-IRSG and the Information Retrieval Journal was also made. As a part of this agreement, starting next year authors of select papers from the Information Retrieval Journal will be invited to present their work at ECIR. Conversely, the award-winning and otherwise top-ranked papers at ECIR will be invited to submit an extended version to the Information Retrieval Journal.
Day 1: Monday, 26th March
The conference started on 26th March, with a mix of six workshops and three tutorials taking place concurrently.
Workshops
- Social Aspects in Personalization and Search (SoAPS)
- Analysis of Broad Dynamic Topics over Social Media (BroDyn)
- Bibliometric-enhanced Information Retrieval: 7th International BIR workshop
- Second International Workshop on Recent Trends in News Information Retrieval (NewsIR’18)
- First International Workshop on Narrative Extraction from Texts (Text2Story 2018)
Being a part of the audience for Text2Story, I was able to witness two very interesting keynotes which approached narrative extraction from very different and equally compelling perspectives. In the first keynote, Udo Kruschwitz talked about how search and browsing patterns of users can be used to improve personalization in enterprise search systems with a limited amount of query logs. Eric Gaussier focused on how the use of word embeddings can be extended with syntactic information and the possible use cases in tasks such as textual entailment.
Tutorials
- Neural Networks for Information Retrieval (NN4IR)
- Extreme Multi-label Classification for Large-scale Text Mining (XMLC-LSTC)
- Semantic Search on Medical Texts (SSMT)
Being delivered by some excellent speakers and the ‘Deep Learning’ Juggernaut continuing to roll, meant the Neural Networks for Information Retrieval (NN4IR) tutorial was the highlight of the day with maximum participation. But, contrary to the general expectation of NN4IR overshadowing other parallel sessions, every workshop and tutorial witnessed a dynamic participation from a good number of attendees. The first day closed with a welcome reception, also held at Minatec. The guests enjoyed the interactions and follow up discussions accompanied by a variety of desserts and wine. The welcome reception ran late into the evening, with the guests in no mood to leave, until being politely asked by the staff to do so.
Day 2: Tuesday, 27th March
The conference began with a welcome note from the organizers followed by the presentation of the Karen Spärck Jones Award to Fernando Diaz. This was then followed by a riveting keynote by the awardee, who spoke about ‘The Harsh Reality of Production Information Access Systems‘. Fernando talked about how most personalization and recommendation systems used in practice today focus on making the majority users happy, while largely ignoring the individuals or users from minority groups. He further highlighted the key challenges that the industry faces in resolving such issues and ended the keynote ended with several open questions. It was followed by a short but intense QA session that was split between some participants supporting a deeper level of personalization and others raising concerns over it.
Post the keynote there were four oral presentation sessions. ‘Topic Modelling’ and ‘Health Applications’ sessions ran in parallel before lunch while those on ‘Deep Learning’ and ‘Evaluation & User behaviour’ took place post lunch. The overall organization of talks was well spaced with plenty of time for lunch and coffee breaks, providing participants to follow up with the speakers or just enjoy conversations and the delicious food. The evening witnessed a very interactive poster and demo session with all the presenters as well as the audience equally enthusiastic. The session also included a public voting for the best poster award. The day officially concluded with Cocktails and an assortment of cheese which provided a much needed relief to the presenters who were starved at the end of a seemingly never-ending poster session.
Day 3: Wednesday, 28th March
The third day of the conference opened with a keynote talk from Gabriella Kazai about ‘Challenges in Building IR Evaluation Pipelines’. The talk focused on various challenges faced in offline and online evaluations of IR systems. Gabriella highlighted how biases introduced during such evaluations can negatively affect the overall system. She went on to share her experience in building IR evaluation pipelines in several roles over the past years. This included designing a new evaluation methodology for focused retrieval task at INEX and sharing her experiences at Microsoft, as well as those at Lumi. Like the previous day, the keynote was followed by four full paper sessions, split into pre and post-lunch sessions. The focus for today was ‘Representation’, ‘Recommendation’, ‘Retrieval’ and ‘Deep Learning’.
The Retrieval session also included a presentation of the paper ‘Statistical Stemmers: A Reproducibility Study’. The authors discussed a comprehensive study conducted on CLEF and TREC collections, to verify the reproducibility of results using three recently proposed statistical stemmers. This work was awarded the ‘Best Paper award at ECIR 2018. The technical sessions were followed by a Panel consisting of Shane Culpepper, Grace Hui Yang, Jochen Leidner and Miguel Martinez and chaired by Nicola Ferro. The Panel debated on the importance of making IR results more interpretable and, going one step further, the utility of being able to predict results of IR systems.
The day ended a little earlier than usual, with the organizing committee arranging three tours for conference participants. The city centre tour turned out to be very dynamic with the guide giving a glimpse into how proud people of Grenoble are of the city’s history. All the tours converged at the restaurant le Téléférique, which hosted the conference banquet. The cable car ride to the top of the hill and the breath-taking view of the city from the top were just a start to the amazing evening. Several awards were announced during the banquet. While the work on ‘Statistical Stemmers: A Reproducibility Study’ by Gianmaria Silvello et al. received the best paper award, Ricardo Campos et al. were voted for the best poster award for “A Text Feature based Automatic Keyword Extraction Method for Single Documents”. ECIR 2018 Test of time award was also announced. Fabrizio Silvestri was chosen for this award for his work on “Sorting out the documents identifier assignment problem”. Ben Carterette et al. were presented the Test of Time Honorable mention award for their paper “Here or There: Preference Judgements for Relevance”.
Another day ended with several participants staying back late into the night until finally being requested to leave. Nonetheless, the party continued late into the night at the pub ‘Shakesbeer’ which, for some reasons, had become popular among the conference participants.
Day 4: 29th March 2018
The final day of the conference started with Radim Řehůřek delivering the Industry day Keynote talk, where he talked about his take on how the research, business and open source communities can work together. Radim raised several important concerns and pointed out the gaps that still exist between the way academic research functions and the actual needs of industry. He also shared his experience and insights from a decade of work in building NLP and ML solutions, and the underlying difficulties he faced when using academic research in a practical setup. Rest of the day was divided between oral presentation sessions and Industry sessions, with both of them running in parallel. The oral sessions on last day covered topics related to ‘Classification’, ‘Micro-blogs’ and ‘Representation’.
The industry session turned out to be a crowd puller in the literal sense, so much so that after the initial sessions organizers decided to shift the remaining sessions to a larger amphitheatre. The session started with a panel, consisting of Julia Kiseleva, Bhaskar Mitra, Craig Macdonald, Agnes van Belle and Radim Řehůřek. The panel was chaired by ‘Gabriella Kazai’. The panel discussed various challenges that are faced when using academic research in production in a commercial setup. This was followed by presentations from candidates from several companies including the likes of ‘Facebook’, ‘Bloomberg’ and ‘myTaxi’, over the day. After three days of ‘academic‘ presentations, it was indeed refreshing to listen to a very different perspective, with most speakers talking about the types of challenges faced in the industry, and how industry and academia can work together in solving them. The day ended with the usual note of thanks and announcement of ECIR 2019 which will take place in Cologne, Germany.