An Extension-Based Argument-Ranking Semantics: Social Rankings in Abstract Argumentation

Lars Bengel, Giovanni Buraglio, Jan Maly, Kenneth SkibaAAAI 2025 · DOI

TL;DR

This paper introduces a novel approach to ranking arguments in abstract argumentation frameworks by combining extension-ranking semantics with social ranking functions. This new method refines existing classifications of arguments (accepted, rejected) into a more nuanced hierarchy, offering a more precise understanding of argument strength within a debate.

What Problem Does This Paper Address?

In abstract argumentation, the quality of an argument is typically assessed by its presence in "extensions," which are sets of jointly acceptable arguments. Standard semantics classify arguments into three categories: skeptically accepted (in every extension), credulously accepted (in at least one extension), and rejected (in no extension). While useful, this tripartite classification lacks the granularity to distinguish between arguments within these categories. For instance, two credulously accepted arguments might have different levels of support or influence, which the current system doesn't capture. To address this, researchers have developed argument-ranking semantics, which aim to provide a more fine-grained ordering. However, these often differ significantly from the more common extension-based classifications. This paper bridges this gap by proposing a method that refines existing extension-based classifications using social ranking functions, creating a more intuitive and technically sound argument-ranking system.

Methods

The core of this paper lies in combining two existing lines of research: extension-ranking semantics and social ranking functions.

  • Extension-Ranking Semantics: These semantics, introduced by Skiba et al. (2021), generalize classical argumentation semantics by providing a ranking over sets of arguments, indicating their relative plausibility of being accepted. The paper utilizes various base relations (like conflict-freeness and defense) to define these extension-rankings.
  • Social Ranking Functions: Originally from economics, these functions rank individuals based on the success of groups they belong to. The paper adapts these to rank arguments based on the quality of the sets they are part of, as determined by the extension-ranking semantics.
  • Generalizing Social Ranking Functions: A key technical contribution is the generalization of social ranking functions to handle partial orders over sets of arguments, using the concept of a "rank of a set."
  • Lexicographic Excellence Operator: The paper specifically employs the lexicographic excellence operator (lex-cel), a prominent social ranking function, to create argument rankings. This operator ranks arguments based on how often they appear in higher-ranked sets, with a lexicographical tie-breaking mechanism.
  • Axiomatic Characterization: The authors define and investigate axioms for social ranking functions and extension-ranking semantics to ensure desirable properties in the resulting argument rankings, particularly a "refinement property" that orders arguments from skeptically accepted to credulously accepted to rejected.

Datasets

Not discussed in the paper. The paper focuses on theoretical frameworks and abstract argumentation frameworks, not empirical evaluation on specific datasets.

Key Insights and Ideas

  • Bridging the Gap: The paper successfully connects the abstract world of extension-based semantics with the more nuanced ordering provided by social ranking functions, creating a unified framework.
  • Refining Argument Quality: The proposed method offers a more granular understanding of argument strength than traditional classifications, allowing for distinctions between arguments that are merely "accepted" or "rejected."
  • Social Ranking for Arguments: The adaptation of social ranking functions from economics to argumentation is a novel and powerful idea, leveraging existing tools to solve a long-standing problem in AI.
  • Lex-cel as a Promising Candidate: The lexicographic excellence operator, when applied to extension-ranking semantics, demonstrates the potential to achieve the desired refinement property, ordering arguments in a way that aligns with intuition about their strength.
  • Axiomatic Foundation: The paper provides a theoretical foundation by identifying necessary and sufficient conditions (axioms) for social ranking functions and extension-ranking semantics to achieve specific desirable properties, such as σ-Compatibility and σ-Refinement.

Key Results

  • The paper defines a new family of argument-ranking semantics based on social ranking functions applied to extension-ranking semantics.
  • It shows that the lexicographic excellence operator, when applied to extension-ranking semantics satisfying σ-generalization, generates an argument ranking that refines the traditional skeptical/credulous/rejected classification. Specifically, skeptically accepted arguments are ranked before credulously accepted, which are ranked before rejected arguments.
  • The paper establishes axiomatic properties for social ranking functions and extension-ranking semantics that are sufficient and necessary for achieving this refinement property.
  • It demonstrates that the lex-cel operator satisfies Pareto-efficiency, a key axiom for ensuring desirable social ranking properties.

Limitations and Future Work

  • Generalization of Social Ranking Functions: While the paper generalizes social ranking functions to handle partial orders, further exploration of other generalized social ranking functions and their properties in argumentation is suggested.
  • Detailed Analysis of Lex-cel: A more in-depth investigation into the properties of the lex-cel argument-ranking semantics, particularly concerning its satisfaction of various principles, is planned for future work.
  • Empirical Evaluation: The paper is theoretical. Future work could involve applying these new semantics to real-world argumentation datasets to evaluate their practical utility and compare them against existing methods.
  • Completeness of Lifting Operators: The paper touches upon "lifting operators" (ranking sets based on object rankings) and notes that existing work in this area for abstract argumentation is not complete, suggesting this as a direction for future research.

Summary updated

BibTeX
@inproceedings{bengel2025extension,
  author = {Lars Bengel and Giovanni Buraglio and Jan Maly and Kenneth Skiba},
  title = {{An Extension-Based Argument-Ranking Semantics: Social Rankings in Abstract Argumentation}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2025},
  doi = {10.1609/aaai.v39i14.33621},
}
Figure