Deliberative AI summarization for online discussions

Our main research question asks : What are the features needed for a synthetic summary to be considered representative and actionable in the case of debates and deliberations ? To address this question, this project will, by proposing summaries with different constructions and features, ask individuals engaging with deliberation.io to rate their perceived views of summaries according to multiple self-reported metrics. Additional to qualitative human judgement, we will use different automatic metrics to evaluate the quantitative qualities of the LLM generated summaries.