Atomic Consistency Preference Optimization for Long-Form Question Answering

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Jingfeng, Thirukovalluru, Raghuveer, Wang, Junlin, Luo, Kaiwei, Dhingra, Bhuwan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914145159348224
author Chen, Jingfeng
Thirukovalluru, Raghuveer
Wang, Junlin
Luo, Kaiwei
Dhingra, Bhuwan
author_facet Chen, Jingfeng
Thirukovalluru, Raghuveer
Wang, Junlin
Luo, Kaiwei
Dhingra, Bhuwan
contents Large Language Models (LLMs) often produce factoid hallucinations - plausible yet incorrect answers. A common mitigation strategy is model alignment, which improves factual accuracy by training on curated (factual, non-factual) pairs. However, this approach often relies on a stronger model (e.g., GPT-4) or an external knowledge base to assess factual correctness that may not always be accessible. Addressing this, we propose Atomic Consistency Preference Optimization (ACPO), a self-supervised preference-tuning method that enhances factual accuracy without external supervision. ACPO leverages atomic consistency signals (i.e., the agreement of individual facts across multiple stochastic responses) to identify high- and low-quality data pairs for model alignment. Despite being fully self-supervised, ACPO outperforms the strong supervised alignment baseline by 1.95 points averaged across Phi-3 and Llama3 on the LongFact and BioGen datasets, demonstrating its effectiveness in improving factual reliability without relying on external models or knowledge bases.
format Preprint
id arxiv_https___arxiv_org_abs_2505_09039
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Atomic Consistency Preference Optimization for Long-Form Question Answering
Chen, Jingfeng
Thirukovalluru, Raghuveer
Wang, Junlin
Luo, Kaiwei
Dhingra, Bhuwan
Computation and Language
Large Language Models (LLMs) often produce factoid hallucinations - plausible yet incorrect answers. A common mitigation strategy is model alignment, which improves factual accuracy by training on curated (factual, non-factual) pairs. However, this approach often relies on a stronger model (e.g., GPT-4) or an external knowledge base to assess factual correctness that may not always be accessible. Addressing this, we propose Atomic Consistency Preference Optimization (ACPO), a self-supervised preference-tuning method that enhances factual accuracy without external supervision. ACPO leverages atomic consistency signals (i.e., the agreement of individual facts across multiple stochastic responses) to identify high- and low-quality data pairs for model alignment. Despite being fully self-supervised, ACPO outperforms the strong supervised alignment baseline by 1.95 points averaged across Phi-3 and Llama3 on the LongFact and BioGen datasets, demonstrating its effectiveness in improving factual reliability without relying on external models or knowledge bases.
title Atomic Consistency Preference Optimization for Long-Form Question Answering
topic Computation and Language
url https://arxiv.org/abs/2505.09039