Token-Level Marginalization for Multi-Label LLM Classifiers

TL;DR

This paper introduces token-level marginalization methods to improve confidence scoring and interpretability of generative language models in multi-label content safety classification, demonstrating significant performance gains.

Contribution

It proposes three novel token-level probability estimation techniques to enhance interpretability and accuracy of LLM classifiers for content safety tasks.

Findings

Token marginalization improves confidence score reliability.

Enhanced interpretability aids fine-grained error analysis.

Framework generalizes across instruction-tuned models.

Abstract

This paper addresses the critical challenge of deriving interpretable confidence scores from generative language models (LLMs) when applied to multi-label content safety classification. While models like LLaMA Guard are effective for identifying unsafe content and its categories, their generative architecture inherently lacks direct class-level probabilities, which hinders model confidence assessment and performance interpretation. This limitation complicates the setting of dynamic thresholds for content moderation and impedes fine-grained error analysis. This research proposes and evaluates three novel token-level probability estimation approaches to bridge this gap. The aim is to enhance model interpretability and accuracy, and evaluate the generalizability of this framework across different instruction-tuned models. Through extensive experimentation on a synthetically generated, rigorously annotated dataset, it is demonstrated that leveraging token logits significantly improves the interpretability and reliability of generative classifiers, enabling more nuanced content safety moderation.