Measuring What Counts: The case of Rumour Stance Classification

TL;DR

This paper critically examines the evaluation metrics used in rumour stance classification, revealing their limitations with imbalanced data, and proposes new metrics that better assess system performance, especially on minority classes.

Contribution

It re-evaluates existing systems from RumourEval and introduces robust metrics tailored for imbalanced rumour stance classification tasks.

Findings

Accuracy and macro-F1 are not reliable for imbalanced data.

Existing metrics favor majority class performance.

Proposed metrics improve evaluation of minority class detection.

Abstract

Stance classification can be a powerful tool for understanding whether and which users believe in online rumours. The task aims to automatically predict the stance of replies towards a given rumour, namely support, deny, question, or comment. Numerous methods have been proposed and their performance compared in the RumourEval shared tasks in 2017 and 2019. Results demonstrated that this is a challenging problem since naturally occurring rumour stance data is highly imbalanced. This paper specifically questions the evaluation metrics used in these shared tasks. We re-evaluate the systems submitted to the two RumourEval tasks and show that the two widely adopted metrics -- accuracy and macro-F1 -- are not robust for the four-class imbalanced task of rumour stance classification, as they wrongly favour systems with highly skewed accuracy towards the majority class. To overcome this problem, we propose new evaluation metrics for rumour stance detection. These are not only robust to imbalanced data but also score higher systems that are capable of recognising the two most informative minority classes (support and deny).