Online petitions are a fascinating object of study. We can learn a great deal from them about how political content “behaves” online. Ironically, the more successful and popular online petitions become, the higher the workload for administrators and operators of such systems. One such issue is petition categorization. Petition systems can leave this up to the users or to their own staff. In both cases, a semi-automatized procedure can prove to be helpful.
In the Digital Governance Systems Lab, Kevin looked into that problem and used data from the UK online petition website to come up with an efficient solution (ACM DL). Demonstrating that this approach can also be applied under the constraint of a legacy online petition system such as in the case in South Korea, Jae Hyun Son had his results presented at ICEDEG 2025 (IEEE Explore). In the case of a multi-label classifier, the problem is trickier. Daniil tackled this challenge in a paper already on the occasion of DGO 2024 in Taipei (ACM DL).
Furthermore, online petitions usually need to pass through an admissibility screening before they can go online. This process requires legal expertise and can become time consuming for administrators as the numbers of petitions increase. Ben approached this topic in two consecutive papers at DGO24 and EGOV25, respectively, with data stemming from the Taiwan JOIN online petition system, adding an element of XAI in the second paper (ACM DL; LNCS).
To further improve the accuracy of language models for downstream tasks, Josie managed to demonstrate that a further pre-trained version of Chinese BERT, CnPBERT, outperforms the standard model for petition categorization (IEEE).
Last but not least, lab members were preparing the grounds for further experiments on the success of online petitions, first learning from qualitative case studies prepared in a paper at ICEDEG 2021 by Jolijne (IEEE) and further exploring petition dynamics by Mate for ICRAI 2024 (ACM). We will continue to work on this strain of research expanding geographically as well as substantially. So stay tuned for more …
Uwe Serdült 