WIBA: What Is Being Argued? A Comprehensive Approach to Argument Mining

TL;DR

WIBA tackles the challenge of comprehensively understanding What Is Being Argued by jointly addressing three interdependent tasks: Argument Detection, Claim Topic Extraction, and Argument Stance Classification. It introduces an ATN-inspired formalization and a data-efficient, prompt-engineered LLM pipeline (WIBA-Detect → WIBA-Extract → WIBA-Stance) that leverages LoRA adapters to fine-tune only lightweight components. The approach yields strong performance across multiple HQ datasets, with $F_1$ scores for detection in the $[80\%,86\%]$ range, $CTE$ scores around $64.8\%$ (and $74.2\%$ on correctly identified arguments), and $F_1$ for stance in the $71\%-78\%$ range, outperforming several baselines and offerings. The work provides an open-access platform at wiba.dev, open-source code and models, and demonstrates practical impact for linguistics, social science, and computer science applications in analyzing large, diverse corpora of argumentative text.

Abstract

We propose WIBA, a novel framework and suite of methods that enable the comprehensive understanding of "What Is Being Argued" across contexts. Our approach develops a comprehensive framework that detects: (a) the existence, (b) the topic, and (c) the stance of an argument, correctly accounting for the logical dependence among the three tasks. Our algorithm leverages the fine-tuning and prompt-engineering of Large Language Models. We evaluate our approach and show that it performs well in all the three capabilities. First, we develop and release an Argument Detection model that can classify a piece of text as an argument with an F1 score between 79% and 86% on three different benchmark datasets. Second, we release a language model that can identify the topic being argued in a sentence, be it implicit or explicit, with an average similarity score of 71%, outperforming current naive methods by nearly 40%. Finally, we develop a method for Argument Stance Classification, and evaluate the capability of our approach, showing it achieves a classification F1 score between 71% and 78% across three diverse benchmark datasets. Our evaluation demonstrates that WIBA allows the comprehensive understanding of What Is Being Argued in large corpora across diverse contexts, which is of core interest to many applications in linguistics, communication, and social and computer science. To facilitate accessibility to the advancements outlined in this work, we release WIBA as a free open access platform (wiba.dev).

Figures (3)

• Figure 1: This figure illustrates the proposed task and methods. The process begins with a given text input, where the first step is to determine the presence of an argument. If an argument exists, the subsequent steps involve extracting the topic being argued, and classifying the stance towards that topic. While each method can operate independently, out of order, Argument Stance Classification requires that a specific topic be provided along with the text.
• Figure 2: Argument Detection Augmented Transition Network (ATN). $\mathcal{S}$ represents the start of the text classification, $\Tilde{\mathcal{A}}$ represents 'Not an Argument', $\mathcal{C}$ represents Claim, $\mathcal{P}$ represents Premise, and $\mathcal{A}$ represents Argument. Looped arrows from $\mathcal{C}$ and $\mathcal{P}$ represent recursive calls to Claim and Premise Augmentation Networks. The symbol $^+$ indicate at least one I.e., $\mathcal{P^+C^+}$ is the state where there is at least one premise with at least one claim.
• Figure 3: Our platform available at https://wiba.dev