Abstract
This paper presents a formal mathematical framework for quantifying the distance between viewpoints in structured debates. By differentiating between agreement with a statement and the perceived importance of an argument, the authors develop a pseudometric for weighted argumentation graphs. This metric is designed to satisfy nine intuitive desiderata for comparing opinions and includes a tunable decay factor to adjust sensitivity to argumentative depth. The framework offers a principled approach for clustering perspectives and facilitating consensus in digital democracy platforms.
Introduction
In the realm of online discourse and digital democracy, understanding the landscape of public opinion is crucial. However, existing tools for comparing viewpoints often rely on simplistic "agree/disagree" metrics that fail to capture the hierarchical nature of human reasoning. Two individuals might reach the same conclusion for entirely different reasons, or they might disagree on a minor point while sharing a fundamental logical framework.
This research addresses the challenge of measuring the distance between complex, structured arguments. The authors propose a mathematical model that distinguishes between agreement (the degree to which one believes a statement is true) and importance (the weight one assigns to a specific argument in reaching a conclusion).
Methodology: Weighted Argumentation Graphs
The approach is built upon the Issue-Based Information System (IBIS) model, where discussions are represented as graphs. Nodes represent statements, and edges represent arguments (pro or con).
Core Components
- Agreement Score ($r$): A value in the range $[-0.5, 0.5]$ representing a user's stance on a statement. A value of 0 indicates neutrality or lack of interest.
- Importance Weight ($w$): A value in the range $[0, 1]$ representing the relative weight of an argument compared to others targeting the same conclusion. For any given node, the weights of all incoming arguments must sum to 1.
The Pseudometric $d_G$
The authors define a distance function, $d_G$, to calculate the difference between two users' graphs. A central feature of this metric is the decay factor $\alpha$. This parameter allows the metric to be tuned: a low $\alpha$ prioritizes top-level conclusions, while a high $\alpha$ increases sensitivity to the underlying "nitty-gritty" details deeper in the argument tree.
Key Insights and Results
The authors identify a critical distinction: the separation of statement rating from argument relevance. This allows the model to capture scenarios where a user might believe a premise is true but considers it irrelevant to the final conclusion.
Mathematical Soundness and Desiderata
The proposed distance measure is formally a pseudometric, satisfying the properties of identity, symmetry, and the triangle inequality. The term "pseudo" is used because if two users have different argument weights but rate all statements as neutral (0), the distance between them is 0. This reflects the intuition that if a user is indifferent to the topic, their underlying logic is practically irrelevant to the comparison.
Furthermore, the authors prove that the metric satisfies nine "desiderata"—logical properties essential for comparing opinions, such as:
- Depth Sensitivity: Deviations in deeper parts of the tree have less influence on the total distance than top-level deviations.
- Polarity Awareness: A "pro" vs. "con" split results in a greater distance than a "pro" vs. "neutral" split.
Limitations and Future Work
While robust, the framework has limitations. The requirement that importance weights sum to 1 leads to normalization side-effects: adding a new argument automatically reduces the relative weight of existing ones, which can shift the calculated distance between users in complex ways.
Additionally, the current model does not explicitly handle "undercuts"—arguments that attack the validity of a logical connection rather than the premise itself. Future work will involve testing the metric in live environments, such as Voting Advice Applications (VAAs), to validate if the mathematical distances align with human perceptions of ideological similarity.
Summary updated