ISWC Trip Report
Last week, the 16th International Semantic Web Conference (ISWC 2017) took place in Vienna, Austria. Around 600 researchers from all over the world came together to exchange knowledge and ideas in 7 tutorials, 18 workshops, and 3 full days of keynotes, conference talks, and a big poster & demo session. Needless to say, I only saw a small part of it, but all the papers and many of the tutorial materials are avaialble through the conference website.
First of all, kudos to the organising committee for putting together a fantastic programme and great overall surroundings. The WU Campus (workshops, posters & demos and jam session) has a really gorgeous campus with a marvellous spaceship-like library.
The main conference took place next door at the Messe, where the Wifi worked excellently (quite a feat at a CS conference where most participants carry more than one device). The bar for next year is set high!
But back to the conference:
On Sunday, I got to present the SERPENS CLARIAH research pilot during the Second Workshop on Humanities in the Semantic Web (WHISE II). There were about 30 participants in the workshop, and a variety of projects and topics was presented. I particularly liked the presentation by Mattia Egloff on his and Davide Picca's work on DHTK: The Digital Humanities ToolKit. They are working on a python module that supports analysis of books and they are developing and testing it for an undergraduate course for humanities students. I really think that by providing (humanities) students with tools to start doing their own analyses, we can get them enthusiastic about programming, as well as thinking about the limitations of such tools, which can lead to better projects in the long run.
In the WHISE workshop, as well as in the main conference, there were several presentations on multimedia datasets for the Semantic Web. The multimedia domain is not new to Semantic Web, but some of the work (such as Rick Meerwaldt, Albert Meroño-Peñuela and Stefan Schlobach. Mixing Music as Linked Data: SPARQL-based MIDI Mashups Mashups) doesn't just focus on the metadata but actually encodes the MIDI signal as RDF and then uses it for a mashup.
Another very interesting resource is IMGpedia, created by Sebastián Ferrada, Benjamin Bustos and Aidan Hogan, which was presented in a regular session (winner best student resource paper) as well as during the poster session (winner best poster). The interesting thing about this resource is that it doesn't only allow you to query on metadata elements, but also on visual characteristics.
Metadata and content features are also combined in The MIDI Linked Data Cloud by Albert Meroño-Peñuela, Rinke Hoekstra, Victor de Boer, Stefan Schlobach, Berit Janssen, Aldo Gangemi, Alo Allik, Reinier de Valk, Peter Bloem, Bas Stringer and Kevin Page which would for example make studies in ethnomusicology possible. I think such combinations of modalities is super exciting for humanities research where we work with extremelty rich information sources and often need to/want to combine sources to answer our research questions.
Enriching and making available cultural heritage data is also a topic that keeps popping up at ISWC, this year there was for example "Craig Knoblock, Pedro Szekely, Eleanor Fink, Duane Degler, David Newbury, Robert Sanderson, Kate Blanch, Sara Snyder, Nilay Chheda, Nimesh Jain, Ravi Raju Krishna, Nikhila Begur Sreekanth and Yixiang Yao: Lessons Learned in Building Linked Data for the American Art Collaborative". This project was a pretty big undertaking in terms of aligning and mapping museum collections. I really like that the first lesson learnt to create reproducible workflows:
This doesn't only hold for conversion of museum collections, but for all research. But it's still nice to see mentioned here. Reproducibility is also a motivator in "Tobias Kuhn, Egon Willighagen, Chris Evelo, Núria Queralt Rosinach, Emilio Centeno and Laura Furlong: Reliable Granular References to Changing Linked Data" which investigates the use of nanopublications to enable referring to items or subsets within data collections for finegrained referencing of previous work.
My favourite keynote at this conference (and they had three excellent ones) was by Jamie Taylor, formerly of Freebase, now Google. He argued for more commonsense knowledge in our knowledge graphs. While I do think that is a great vision, as many of our resources lack this leading to all sorts of weird outcomes in for instance named entity linking (you can ask Filip Ilievski for the funniest examples) it was unclear how to go about this this and whether this would be possible at all. The examples he gave in the keynote for toasters and kettles would work out just fine (kettles heat up water, toasters heat up baked goods) but for complex concepts such as murders (Sherlock Holmes anyone?) I'm not sure how this would work. But enough food for thought. See also Pascal Hitzler's take on this keynote.
See you in Monterey, California next year?