A two-day conference was held at the headquarters of the European Broadcasting Union (EBU) in Geneva on 8 and 9 November. I had the pleasure of hosting a workshop on ethics and artificial intelligence aspects on the first day and giving an overview of the RecSys 2018 conference on the second day. In the latter, I presented what interested me most about RecSys 2018 (the most renowned conference on recommendation systems) which was held this year in Vancouver, Canada. The plan of my presentation was structured around the following themes:
RecSys 2018 in figures
The RecSys 2018 conference attracted 817 participants this year, 73% of whom were practitioners and 27% academics. It is a unique distribution that makes for a very enriching conference because it is at the crossroads of both worlds.
2 tutorials not to be missed
I followed two informative and interesting tutorials.
Mixed methods for evaluating recommendations
An incredibly rewarding workshop that made me realise that market research techniques are advantageous even in complex and highly technical contexts such as recommendation systems.In a nutshell: use qualitative research to build your model and collect quantitative data to verify it. It’s obvious, but I failed to recognise that in some cases you need to go back to the basics. What may seem evident can sometimes appear distant.
Personality and recommendations
I have already made a detailed summary of this tutorial in another article that I recommend you read. The main idea was to raise participants’ awareness of the role played by personality in interacting with referral systems.
3 remarkable keynotes
The 3 keynotes (one per day) were given by Elisabeth Churchill, Lise Getoor (an overview on the subject of Probabilistic Soft Logic or “PLS” in short) and Christopher Berry (CBC) on the role of the recommendation in social cohesion. The 3 keynotes were all fantastic, but I must admit that Christophe Berry’s, as an excellent storyteller, had a number of them taped.
Some remarkable research papers (among many others)
Of course, it was impossible to summarise an entire conference and the more than 50 research papers presented in a few minutes. I have, therefore, chosen to illustrate the quality of RecSys through 3 works that particularly appealed to me and were related to the audio-visual sector:
- a series of experiments on the controllability of a Spotify interface
- a very innovative analysis of the measurement of inaction (one of my favourite articles that I have discussed in detail in this post)
- artworks personalisation on the Netflix interface
2 workshops: FATRec and IntRS
The last two days of the conference were devoted to workshops.
- this year I had the pleasure of organising FATRec with my colleagues Toshihiro Kamishima and Michael Ekstrand.
- and I attended the excellent IntRS workshop on human-machine interfaces, during which I learned a lot (thanks in particular to Eric Vorm’s presentation on the Q method)
If you would like to take a look at my presentation, I have made it available below. Feel free to ask any questions and share it if you think it might be useful.Tags: media, recommendation algorithms