PAKT: Perspectivized Argumentation Knowledge Graph and Tool for Deliberation Analysis (with Supplementary Materials)

TL;DR

PAKT introduces a Perspectivized Argumentation Knowledge Graph to analyze deliberation by structurally organizing arguments into premises, conclusions, frames, values, and background knowledge. It constructs PAKTDDO from debate.org, automatically generates conclusions, and enriches arguments with frame and value annotations grounded in ConceptNet, enabling cross-issue and camp-level analytics. The framework is demonstrated through global analyses and case studies (notably animal hunting debates), revealing how frames, values, and concepts shape deliberation across stakeholder camps and issues, and supporting advanced retrieval of similar or counter-arguments. Public deployment via a Neo4J backend and a web interface provides a practical tool for researchers and policymakers to explore deliberation dynamics, while acknowledging data-noise and evaluation limitations as areas for future refinement.

Abstract

Deliberative processes play a vital role in shaping opinions, decisions and policies in our society. In contrast to persuasive debates, deliberation aims to foster understanding of conflicting perspectives among interested parties. The exchange of arguments in deliberation serves to elucidate viewpoints, to raise awareness of conflicting interests, and to finally converge on a resolution. To better understand and analyze the underlying processes of deliberation, we propose PAKT, a Perspectivized Argumentation Knowledge Graph and Tool. The graph structures the argumentative space across diverse topics, where arguments i) are divided into premises and conclusions, ii) are annotated for stances, framings and their underlying values and iii) are connected to background knowledge. We show how to construct PAKT and conduct case studies on the obtained multifaceted argumentation graph. Our findings show the analytical potential offered by our framework, highlighting the capability to go beyond individual arguments and to reveal structural patterns in the way participants and stakeholders argue in a debate. The overarching goal of our work is to facilitate constructive discourse and informed decision making as a special form of argumentation. We offer public access to PAKT and its rich capabilities to support analytics, visualizaton, navigation and efficient search, for diverse forms of argumentation.

Figures (7)

• Figure 1: PAKT data model consisting of arguments (w/ premises, conclusions, frames, values, stance towards topic and concepts) and authors, camps, zeitgeist
• Figure 2: Correlation between frames and values. Left plot is across all topics, right plot is for the issue Should animal hunting be banned? Arguments labeled with more than one frame/value are counted multiple times. Numbers are percentages.
• Figure 3: Comparison between frame and value matrices. The left and middle plots show distributions in percent, and the right plots show their differences in percentage points (pp).
• Figure 4: T-SNE embedding of the spectral embeddings of the largest connected component of the friendship network of DDO. Users replying to Should animal hunting be banned? ($\star$), Should animal testing be banned? ($\bullet$) or Should humans stop eating animals and become vegetarians? ($\bm+$) are marked in blue (pro) or red (con). We see that camps are embedded consistently across similar issues.
• Figure 5: Screenshot of an opinion poll on debate.org