ProbPlug improves confidence estimates for large language model classifiers
ProbPlug: A Plugin Uncertainty Network for Reliable Confidence in LLM Binary Classification
Computation and Language
Summary
Large language models can classify text well, but sometimes they are unsure if their answers are right. The authors created ProbPlug, a tool that checks how confident the model should be by looking inside the model's own thinking process. ProbPlug helps the model know when it might be wrong without changing the original model. Tests show ProbPlug gives better confidence scores and keeps predictions reliable across many tasks.
What this means in practice
- •For machine learning engineers: Add reliable confidence measures to LLM-based classifiers to improve trustworthiness in critical decision systems.
- •For software developers: Integrate confidence checks into existing LLM-based apps without modifying the main model for better user feedback.
Authors
Jianzong Wang, Chuhang Liu, Botao Zhao, Zuheng Kang, Xulong Zhang, Xiaoyang Qu, Junqing Peng, Zhiewei Ye, Yayun He
Abstract
Large language models (LLMs) have achieved strong performance across a broad range of classification settings, yet the reliability of their predictions remains a major obstacle to deployment in high-stakes scenarios. Although confidence estimation for LLMs has been widely studied, confidence calibration for LLM-based classification remains underexplored. We introduce ProbPlug, a lightweight confidence estimation framework for LLM-based binary classification, which predicts whether an output is correct using internal token features extracted from a frozen LLM. ProbPlug employs a self-attention module to aggregate hidden representations and can be integrated into the original inference pipeline without modifying the base model. Experiments across multiple tasks involving both text-based and multimodal large models show that ProbPlug provides more reliable confidence estimates, improves classification performance with negligible additional overhead, and exhibits strong generalization across tasks. These results indicate that ProbPlug serves as a practical solution for confidence estimation in LLM-based classification. Our code is publicly available at Github.