Truth or Twist? Optimal Model Selection for Reliable Label Flipping Evaluation in LLM-based Counterfactuals

TL;DR

This paper investigates how different relationships between models affect the reliability of label flipping evaluation in counterfactual data augmentation for large language models, highlighting the importance of model independence.

Contribution

It systematically analyzes the impact of various model relationships on evaluation reliability and identifies independent judge models as most effective.

Findings

Independent judge models yield the most reliable label flipping evaluations.

Relationships aligned with user preferences improve model performance and robustness.

A significant gap remains between automated evaluations and human judgment, indicating need for human intervention.

Abstract

Counterfactual examples are widely employed to enhance the performance and robustness of large language models (LLMs) through counterfactual data augmentation (CDA). However, the selection of the judge model used to evaluate label flipping, the primary metric for assessing the validity of generated counterfactuals for CDA, yields inconsistent results. To decipher this, we define four types of relationships between the counterfactual generator and judge models: being the same model, belonging to the same model family, being independent models, and having an distillation relationship. Through extensive experiments involving two state-of-the-art LLM-based methods, three datasets, four generator models, and 15 judge models, complemented by a user study (n = 90), we demonstrate that judge models with an independent, non-fine-tuned relationship to the generator model provide the most reliable label flipping evaluations. Relationships between the generator and judge models, which are closely aligned with the user study for CDA, result in better model performance and robustness. Nevertheless, we find that the gap between the most effective judge models and the results obtained from the user study remains considerably large. This suggests that a fully automated pipeline for CDA may be inadequate and requires human intervention.