Adaptive Questionnaires Paper

Publication in ECML/PKDD Conference

We’re excited to announce that our submission was accepted as a paper in the European Conference for Machine Learning (ECML) . Fynn Bachmann’s work on “Fast and Adaptive Questionnaires for Voting Advice Applications ” aims at making data collection and political surveys more effective by using item response theory and computerised adaptive testing. To demonstrate this approach, he uses data from the Swiss voting advice application Smartvote and simulates different question selection policies to compare their speed of convergence.

Over the summer, this work had already been often presented in several conferences and workshops. We were glad to participate in the workshop “Exploring the Role of AI in Deliberation” in Zurich in May 2024. There, we presented the preprint in combination with the overall D3-Project. Other speakers were Francesco Veri and James Fishkin with whom we were happy to exchange ideas and feedback.

Furhtermore, the work in progress got accepted to the 10th International Conference for Computational and Social Science (IC2S2) . The community showed great interest in our approach, when we presented the paper at the conference held in Philadelphia, USA. The focus here was more on the interdisciplinary approach of developing statistical models for political science data, our next presentation at the General Conference for the European Consortium for Political Research (ECPR) targeted practical applications. In the session on voting advice applications, we got in touch with researchers and developers of voting advice applications, with whom many potential collaborations were discussed. It was inspiring to see so many interesting talks across four different panels on the same topic.

Having submitted this work already in March 2024, the paper then got accepted at the European Conference for Machine Learning (ECML) as a full paper publication. Despite having presented the work already at many different venues, it was still an exciting experience to finally present it for a machine learning community with its very technical and optimisation oriented focus. Fynn Bachmann gave talks both at the Ph.D. symposium and the panel for Recommender Systems, and presented a poster in the poster session. Our participation at the conference held in Vilnius, Lithuania, was a great success.

The work is now published in the proceedings of the ECML conference and available at the link on Springers website . For people without subscription, the pre-print is still available on ArXiv .