ANCHOR: Abductive Network Construction with Hierarchical Orchestration for Reliable Probability Inference in Large Language Models

2026-05-11

AI Summary

This work addresses the challenges faced by large language models in large-scale decision-making under incomplete information, where they often produce “unknown” predictions and are adversely affected by factor sparsity and spurious correlations. To mitigate these issues, the authors propose a novel approach that integrates hierarchical factor spaces with causal Bayesian inference. The method iteratively generates and clusters factors to construct a dense, ordered factor structure, and employs a context-aware hierarchical retrieval-and-refinement mechanism to enable more accurate probabilistic mapping. Furthermore, a causal Bayesian network is embedded within a naive Bayes framework to relax its overly strong conditional independence assumption. The proposed approach significantly reduces “unknown” predictions, enhances the reliability of probability estimates, achieves state-of-the-art performance, and substantially lowers computational and token costs.

Abstract

A central challenge in large-scale decision-making under incomplete information is estimating reliable probabilities. Recent approaches leverage Large Language Models (LLMs) to generate explanatory factors and elicit coarse-grained probability estimates. Typically, an LLM performs forward abduction to propose factors, each paired with two mutually exclusive attributes, and a Naïve Bayes model is trained over factor combinations to refine the final probabilities. However, sparse factor spaces often yield ``unknown'' outcomes, while expanding factors increases noise and spurious correlations, weakening conditional independence and degrading reliability. To address these limitations, we propose \textsc{Anchor}, an inference framework that orchestrates aggregated Bayesian inference over a hierarchically structured factor space. \textsc{Anchor} first constructs a dense and organized factor space via iterative generation and hierarchical clustering. It then performs context-aware mapping through hierarchical retrieval and refinement, substantially reducing ``unknown'' predictions. Finally, \textsc{Anchor} augments Naïve Bayes with a Causal Bayesian Network to capture latent dependencies among factors, relaxing the strict independence assumption. Experiments show that \textsc{Anchor} markedly reduces ``unknown'' predictions and produces more reliable probability estimates than direct LLM baselines, achieving state-of-the-art performance while significantly reducing time and token overhead.

Problem

• probability inference
• large language models
• abductive reasoning
• factor space
• conditional independence

Innovation

• Hierarchical Factor Space
• Causal Bayesian Network
• Abductive Reasoning
• Reliable Probability Inference
• Large Language Models