---
id: 20260821-T0-03
title: "LongNovel新基准：专测大模型长篇小说摘要的幻觉问题"
title_en: "New Benchmark LongNovel Targets Hallucinations in Long-Novel Summarization"
url: https://ai.daily.yangsir.net/daily/20260821-T0-03
issue_date: 2026-08-21
publish_date: 2026-08-20T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2608.18082
---

# LongNovel新基准：专测大模型长篇小说摘要的幻觉问题

arXiv 研究发布了 LongNovel，一个专门用于检测大模型在长篇novel摘要任务中产生幻觉的多尺度基准。研究指出，尽管上下文窗口越来越大，但长上下文摘要中的幻觉问题依然严峻。相比新闻或论文，长篇小说因信息密度高，更能有效暴露这一缺陷。该基准为评估和改善模型在这一特定任务上的可靠性提供了新工具。

## English Version

**New Benchmark LongNovel Targets Hallucinations in Long-Novel Summarization**

Researchers introduce LongNovel, a multi-scale benchmark designed specifically for detecting hallucinations in long-context novel summarization. The study highlights that while context windows have grown, hallucination remains a severe challenge. Novels, with their high intrinsic information density, are argued to be better suited than news for exposing these issues, providing a new tool for evaluating model reliability.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2608.18082)

**详情页**：https://ai.daily.yangsir.net/daily/20260821-T0-03

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*