publications

2026

1. ACM CUI

   

   Towards a Typology of User Engagement in Conversational Agent Voting Advice Applications

   Daan Van Der Weijden, Thilo Iganz Dieing, and Fynn Bachmann

   In Proceedings of the 2026 ACM Conference on Conversational User Interfaces (CUI ’26)

   Authors contributed equally

   Honorable Mention Abs DOI

   Honorable Mention Poster Award at the 2026 ACM Conference on Conversational User Interfaces (CUI ’26), Bremen, Germany.

   Voting Advice Applications (VAAs) help citizens align with political parties, but are limited by frequent comprehension problems. Conversational Agent VAAs (CAVAAs) address this by integrating chatbot-based support. Yet, user interaction patterns and their effects on completing the CAVAA remain underexplored. This study identifies behavior-based CAVAA user types and examines their interaction with chatbot personas. Using interaction data from 189 users of an GPT-driven CAVAA during the 2024 European Parliament elections, a Latent Class Analysis reveals three types: Checkers (low interaction), Seekers (high engagement and uncertainty), and Testers (system probing rather than advice seeking). While user types do not predict completion, the chatbot personas significantly did. We find that the more active chatbot (asking follow-up questions) increased dropout rates. Our analysis introduces a novel behavioral typology and highlights the importance of conversational design for reducing dropout and improving CAVAA effectiveness.

2. Pol. and Gov.

   

   Estimating the Recommendation Certainty in Candidate-based Voting Advice Applications

   Fynn Bachmann, Daan Van Der Weijden, Cristina Sarasua, and Abraham Bernstein

   Politics and Governance, Jan 2026

   Abs DOI

   Voting Advice Applications (VAAs) typically require users to answer questionnaires before receiving party or candidate recommendations. As users answer more questions, the recommendations naturally become more accurate. However, when users do not complete the questionnaire, the certainty of these recommendations is unknown. In this work, we develop and present a measure to quantify this certainty by introducing an algorithm that estimates the Candidate Recommendation Accuracy (CRA) – the overlap between early and final recommendations – after each question. Through simulations based on existing voter data, we find that our algorithm is more accurate than heuristic estimates. Additionally, it can identify stable recommendations – candidates who are likely to be among the final recommendations – with fewer false positives. Furthermore, we conduct a user experiment investigating different ways of communicating recommendation certainty to users. Our results show that users answer more questions when they see a preview of stable recommendations, but quit the questionnaire earlier when we display an artificially high CRA estimate. Moreover, we find that users appreciate the interface’s simplicity over its accuracy. We conclude that displaying personalized stable recommendations can spark curiosity in VAAs while providing a robust estimate of recommendation certainty for users who submit incomplete questionnaires.

3. PyPI

   IXPLORE: Bounded Ideal Point Estimation with Grid-based Uncertainty Quantification

   Fynn Bachmann

   2026

   Python package, version 1.2

   Abs arXiv Code

   Ideal point estimation is widely used to analyze and visualize political data. However, selecting the corresponding spatial model involves various trade-offs: while Item Response Theory (IRT) approaches are based on utility functions rather than optimized for predictive accuracy, most Machine Learning (ML) alternatives struggle to generalize beyond training data when embedding sparse test responses. We introduce IXPLORE, a bounded ideal point estimation algorithm that combines a predictive fit objective with a sparsity-aware likelihood function. On five benchmark datasets spanning surveys, roll calls, and deliberation, this approach surpasses IRT- and ML-based algorithms on reconstruction and imputation error – especially for users with sparse responses. Furthermore, we show that non-linear feature transforms can further reduce the reconstruction error while remaining visually interpretable. To quantify uncertainty, IXPLORE applies grid-based posterior inference on a bounded 2D latent space. Available as a Python package on PyPI, IXPLORE offers a flexible framework for constructing bounded, interpretable political maps with fast inference and strong imputation performance.

4. working paper

   

   User-centered Evaluation of Adaptive Questionnaires in Voting Advice Applications: Insights from two Online Experiments on the AQVAA Platform

   Fynn Bachmann, Cristina Sarasua, and Abraham Bernstein

   Working paper, under review at the Journal of Information Technology & Politics

   Abs Code

   Adaptive Questionnaires (AQs) can improve the recommendation process in Voting Advice Applications (VAAs) by selecting the most informative next question based on users’ individual response profiles. While the statistical methods to achieve this information gain are well studied, fewer studies focus on the user perspective within AQs. To address this research gap, we conduct two user experiments on the AQVAA platform, an online VAA that implements several adaptive selection strategies. First, we evaluate how users assess candidate recommendations and find that highly engaged users are more satisfied with recommendations in AQs than in equally long, static questionnaires. Second, we evaluate how users perceive the adaptivity of the system and find that some users appreciate the transparency of the question selection process, although many users have difficulties understanding the underlying logic. This evidence suggests that AQs can improve the user experience in VAAs, while further adjustments are needed to explain their rationale transparently. By providing the AQVAA platform as open-source software, this exploratory study prepares future user experiments on the enhancement of political education with AQs.

2025

1. ACM FAccT

   

   Bridging Voting and Deliberation with Algorithms: Field Insights from vTaiwan and Kultur Komitee

   Joshua C. Yang and Fynn Bachmann

   In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency

   Presented at FAccT 2025 in Athens, Greece

   Abs DOI

   Democratic processes increasingly integrate large-scale voting with face-to-face deliberation to reconcile individual preferences with collective decision-making. This work introduces algorithmic methods to bridge online voting with face-to-face deliberation, tested in two real-world scenarios: Kultur Komitee 2024 (KK24) and vTaiwan. We present three key contributions: (1) Preference-based Clustering for Deliberation (PCD), enabling both focused and broad discussions by computing balanced homogeneous and heterogeneous groups; (2) Humanin-the-loop MES, enhancing the Method of Equal Shares algorithm with real-time feedback, giving participants control over algorithmic decision-making; and (3) the ReadTheRoom method, using opinion mapping to identify agreement and divergence while tracking opinion shifts during deliberation. These actionable frameworks extend in-person deliberation with scalable digital methods that address the complexities of modern participatory decision-making.

2. PLoS One

   

   Adaptive political surveys and GPT-4: Tackling the cold start problem with simulated user interactions

   Fynn Bachmann, Daan Van Der Weijden, Lucien Heitz, Cristina Sarasua, and Abraham Bernstein

   PLoS One, May 2025

   Abs DOI

   Adaptive questionnaires dynamically select the next question for a survey participant based on their previous answers. Due to digitalisation, they have become a viable alternative to traditional surveys in application areas such as political science. One limitation, however, is their dependency on data to train the model for question selection. Often, such training data (i.e., user interactions) are unavailable a priori. To address this problem, we (i) test whether Large Language Models (LLM) can accurately generate such interaction data and (ii) explore if these synthetic data can be used to pre-train the statistical model of an adaptive political survey. To evaluate this approach, we utilise existing data from the Swiss Voting Advice Application (VAA) Smartvote in two ways: First, we compare the distribution of LLM-generated synthetic data to the real distribution to assess its similarity. Second, we compare the performance of an adaptive questionnaire that is randomly initialised with one pre-trained on synthetic data to assess their suitability for training. We benchmark these results against an “oracle” questionnaire with perfect prior knowledge. We find that an off-the-shelf LLM (GPT-4) accurately generates answers to the Smartvote questionnaire from the perspective of different Swiss parties. Furthermore, we demonstrate that initialising the statistical model with synthetic data can (i) significantly reduce the error in predicting user responses and (ii) increase the candidate recommendation accuracy of the VAA. Our work emphasises the considerable potential of LLMs to create training data to improve the data collection process in adaptive questionnaires in LLM-affine areas such as political surveys.

2024

1. ECML/PKDD

   

   Fast and Adaptive Questionnaires for Voting Advice Applications

   Fynn Bachmann, Cristina Sarasua, and Abraham Bernstein

   In Machine Learning and Knowledge Discovery in Databases. Applied Data Science Track

   Presented at ECML 2024 in Vilnius, Lithuania

   Abs DOI

   The effectiveness of Voting Advice Applications is often compromised by the length of their questionnaires. To address user fatigue and incomplete responses, some applications (such as the Swiss Smartvote) offer a condensed version of their questionnaire. However, these condensed versions cannot ensure the accuracy of recommended parties or candidates, which we show to remain below 40%. To address these limitations, this work introduces an adaptive questionnaire approach that selects subsequent questions based on users’ previous answers, aiming to enhance recommendation accuracy while reducing the number of questions posed to the voters. Our method uses an encoder and decoder module to predict missing values at any completion stage, leveraging a two-dimensional latent space that is reflective of the traditional methods used in political science for visualizing ideology. Additionally, a selector module is proposed to determine the most informative subsequent question based on the voter’s current position in the latent space and the remaining unanswered questions. We validated our approach using the Smartvote dataset from the Swiss Federal elections in 2019, testing various spatial models and selection methods to optimize the system’s predictive accuracy. Our findings indicate that employing the IDEAL model both as encoder and decoder, combined with a PosteriorRMSE method for question selection, significantly improves the accuracy of recommendations, achieving 74% accuracy after asking the same number of questions as in the condensed version.

2023

1. ECML/PKDD

   

   Wasserstein t-SNE

   Fynn Bachmann, Philipp Hennig, and Dmitry Kobak

   In Machine Learning and Knowledge Discovery in Databases

   Presented at ECML 2022 in Grenoble, France

   Abs DOI

   Scientific datasets often have hierarchical structure: for example, in surveys, individual participants (samples) might be grouped at a higher level (units) such as their geographical region. In these settings, the interest is often in exploring the structure on the unit level rather than on the sample level. Units can be compared based on the distance between their means, however this ignores the within-unit distribution of samples. Here we develop an approach for exploratory analysis of hierarchical datasets using the Wasserstein distance metric that takes into account the shapes of within-unit distributions. We use tSNE to construct 2D embeddings of the units, based on the matrix of pairwise Wasserstein distances between them. The distance matrix can be efficiently computed by approximating each unit with a Gaussian distribution, but we also provide a scalable method to compute exact Wasserstein distances. We use synthetic data to demonstrate the effectiveness of our Wasserstein t-SNE, and apply it to data from the 2017 German parliamentary election, considering polling stations as samples and voting districts as units. The resulting embedding uncovers meaningful structure in the data.