Charlie Hull
Category: Conference
Two weeks ago we ran our first ever virtual relevance conference – Haystack 2020, as part of a combined week of events including Berlin Buzzwords and MICES, the e-commerce search event. This was a new experience for everybody, including the organising team – so did it work?
Planning an entirely new type of event was always going to be a challenge, but we were lucky enough to be working with an extremely dedicated team from Plain Schwartz (who had helped us run Haystack Europe 2019 and are the team behind the annual Berlin Buzzwords conference). René Kriegler and Paul Bartusch, old friends of OSC and the organisers of MICES (the Mix-Camp for Ecommerce Search, usually held in Berlin the day after Buzzwords) also joined the team. During our weekly calls leading up to the event we wondered who would agree to speak, how many people would attend, what time zones to use, how to run interactive workshops effectively…the list of questions was endless. Nina Müller has written a great blog about how it all came together from the organiser’s perspective, but I’m mainly going to concentrate on the search-related talks and sessions, give some highlights and what we learned from the week. I’m very pleased to say over 250 people were involved in the combined conference.
AMA on AI
Most of the Haystack and MICES sessions ran from Wednesday to Friday, but on Monday night I was lucky enough to moderate an Ask me Anything session on AI-Powered Search with OSC’s Doug Turnbull and Trey Grainger (until very recently Chief Algorithms Officer at Lucidworks but now striking out on his own with his new company Searchkernel). Doug and Trey are working on a new book for Manning on this subject so this was a great chance to find out their views on what’s coming next – and whether it will challenge the dominance of text-based retrieval. I think the overall answer was ‘maybe’ – but hybrid approaches, incorporating the best of both old and new techniques (like vector search) are more likely to be effective.
Unfortunately I didn’t catch any of the sessions on Tuesday, although the Lucene 9 talk from Uwe Schindler (who knows more about the low-level internals of Lucene than pretty much anyone I know) would have been fun to attend. The questions posted for Uwe in Slack looked fascinating (an example issue was how Lucene might take advantage of GPUs for processing which led me to this interesting JIRA issue – still a lot of work to be done it seems!). Vector-based search was also discussed. That evening there was a virtual get-together in a virtual bar!
A Chorus at MICES
Wednesday opened with our joint Haystack/MICES introduction and then a series of talks on e-commerce search. I enjoyed hearing how Carrefour had moved from a commercial search solution to open source ‘owned’ by their team – which is certainly something we’d recommend. Zalando’s talk on Neural Search was fascinating – it’s clear that some of the larger e-commerce companies are really pushing the boundaries of what’s possible and taking advantage of new techniques. The next talk from Fabien Engeln was more on strategy and showed his company’s Exploded View model as applied to e-commerce search.
The day continued with a joint talk from OSC’s Eric Pugh, René Kriegler, Paul Bartusch and Johannes Peter on something they had been cooking up for a few weeks before the event – a complete demonstration stack of a web shop based on open source technologies named Chorus, with built-in tools for measuring and tuning search relevance. The idea is to offer an alternative route to using a commercial e-commerce platform – yes, again this is a way to own your search! We’re going to be writing more about Chorus over the next few weeks – you’ll be meeting Pete the Product Owner who has some challenges ahead of him!
Haystacking lessons
Thursday featured the start of our Haystack programme – all the talks were drawn from those selected by our CFP process for the original conference (note that there’s more Haystack talks happening in our weekly Haystack LIVE! Meetup). Jettro and Byron of Luminis gave some great insights into the lessons they’ve learned from over a decade working in search (there were a few quiet groans of recognition from me, but luckily virtual events have mute buttons). Erica Lesyshyn of EBSCO showed us how to use taxonomies to add related terms and a great tool used for relevance judgement gathering in the medical domain, and Peter Dixon-Moses showed some practical ways to use click log data to improve relevance. We finished the day with a panel discussion on result diversity in search introduced by Doug Turnbull, featuring three different perspectives from Andreas Wagner, Felipe Besson and Lev Gershenzon. We’d had a number of talks submitted to MICES and Haystack on the subject and thought it best to bring everyone together to offer their views.
Friday was the last day of Haystack talks and of the combined conference. Trey Grainger returned to give his views on Thought Vectors and other cutting-edge new techniques; Tim Allison showed us how genetic algorithms can be used to evolve the ‘best’ settings for the myriad configuration options for modern search engines and Jacopo Tagliabue talked about new personalization techniques using deep learning. As during the whole week, all our speakers were available to take questions after their talk and in a breakout room afterwards – and thanks to an engaged audience we never lacked for questions!
Never the same again?
As I said during a brief organiser’s closing presentation, we never came up with a great name for the combined event – BuzzHayMice? HayBuzzMice? MiceBuzzingInAHaystack? – but I think that considering the circumstances we ended up with a great programme of talks, panels and discussions. I also wondered if some future events might be not just online, not just in person, but a hybrid – making it possible for more people in more places to participate? I don’t think things are ever going to be quite the same again for conferences – maybe they shouldn’t be.
Haystack will return – but in the meantime, the videos of all the sessions are gradually appearing online – we’ll add the Haystack talks to our YouTube channel and you can find the rest on Plain Schwartz’ channel. Thanks again to everyone who believed we could make this happen and supported the event – it wouldn’t have been possible without you.