The next speaker in this Weizenbaum Conference session is Quentin Bukold, whose interest is in mass comment campaigns, for instance in response to European Union public consultation efforts. Such campaigns encourage large numbers of supporters to send a pre-formulated text in response to an online consultation effort; this is essentially spamming the consultation form, but might nonetheless represent some facet of public opinion.
Mass comment campaigns thereby jeopardise legitimate public consultation processes, but also provide information about the mobilisation potential of critical interest groups; however, there are few effective ways for identifying, describing, and responding to such campaigns.
This hinges on making the consultation data usable; Quentin scraped a dataset of some 500,000 multilingual comments from EU public consultation efforts, and a sought to cluster these comments by text similarity (using the Jaccard score). The messages promoted by these clusters were then summarised using Large Language Models; these summaries were then further validated for accuracy. Comment clusters were then also correlated with their timestamps, to examine the temporal evolution and burstiness of these comment campaigns.
As it turns out, mass comment campaigns sometimes dominate public consultation efforts; a large number of contributions does not always translate to representative participation, therefore. Comment texts are often also slightly modified by their posters, making detection more difficult. A lot more needs to be done to identify such campaigns and examine their effects on legislation.
