FactReasoner: A Probabilistic Approach to Long-Form Factuality Assessment for Large Language Models

2025-02-25

AI Summary

Large language models (LLMs) often generate factually unreliable content, limiting their deployment in high-stakes, precision-critical applications. To address this, we propose FactReasoner—a probabilistic reasoning framework for factual assessment of long-form generated text. It decomposes model outputs into atomic propositions, retrieves external evidence via knowledge retrieval, leverages pretrained language models to identify semantic entailment or contradiction between each proposition and its supporting context, and constructs a joint probabilistic graphical model to infer proposition-level factual support scores via Bayesian posterior inference. Crucially, FactReasoner is the first method to explicitly encode logical semantic relations as differentiable probability distributions, enabling end-to-end, interpretable fact verification—moving beyond heuristic prompting and binary classification paradigms. Experiments demonstrate that FactReasoner significantly outperforms state-of-the-art prompting-based approaches on both supervised and unsupervised benchmarks, achieving simultaneous gains in factual precision and recall, thereby validating the efficacy and generalizability of probabilistic modeling for factual evaluation.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Knowledge Representation and Reasoning: Reasoning with Beliefs

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large language models (LLMs) have demonstrated vast capabilities on generative tasks in recent years, yet they struggle with guaranteeing the factual correctness of the generated content. This makes these models unreliable in realistic situations where factually accurate responses are expected. In this paper, we propose FactReasoner, a new factuality assessor that relies on probabilistic reasoning to assess the factuality of a long-form generated response. Specifically, FactReasoner decomposes the response into atomic units, retrieves relevant contexts for them from an external knowledge source, and constructs a joint probability distribution over the atoms and contexts using probabilistic encodings of the logical relationships (entailment, contradiction) between the textual utterances corresponding to the atoms and contexts. FactReasoner then computes the posterior probability of whether atomic units in the response are supported by the retrieved contexts. Our experiments on labeled and unlabeled benchmark datasets demonstrate clearly that FactReasoner improves considerably over state-of-the-art prompt-based approaches in terms of both factual precision and recall.

Problem

• Assessing factual accuracy in LLM-generated content
• Decomposing long-form responses for probabilistic evaluation
• Improving precision and recall in factuality assessments

Innovation

• Probabilistic reasoning for factuality
• Decomposition into atomic units
• Joint probability distribution construction