Online debates offer an unprecedented amount of information about controversial issues, bringing together diverse arguments and viewpoints in a common space. However, the growing size and complexity of these debates make it difficult for individuals to fully explore and assess them. Moreover, the information available for evaluating individual arguments is often incomplete, as many contributions receive little or no feedback. These limitations raise the question of how meaningful and reliable assessments of controversial issues can be obtained when both the exploration of a debate and the information available about its arguments are incomplete or uncertain.
This research investigates this question through computational argumentation, with a particular focus on quantitative bipolar argumentation frameworks. It studies how partial reading of large-scale online debates affects their assessment and develops methods for representing and propagating uncertainty in the evaluation of arguments. The proposed approaches combine formal modelling, uncertainty representation, and empirical analysis of real-world online debates. Together, they aim to make complex structures of disagreement more tractable while preserving the uncertainty and limitations inherent in the available information, ultimately supporting a more robust understanding of controversial issues.