arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2308.00399cs.CLcs.LG
收藏

解决神经图表摘要中的幻觉问题

Tackling Hallucinations in Neural Chart Summarization

Charles University(查理大学);Saarland University(萨尔大学);Julius-Maximilians-Universität Würzburg(维尔茨堡朱利叶斯-马克西米利安大学)

Saad Obaid ul Islam, Iza Škrjanec, Ondřej Dušek, Vera Demberg

AI总结:

本文针对神经图表摘要中的幻觉问题,提出基于自然语言推理的训练数据预处理方法,并通过缩短长距离依赖和引入标题、图例等图表信息提升摘要性能。

摘要:

文本生成中的幻觉是指系统生成的文本缺乏输入依据。本文针对神经图表摘要中的幻觉问题展开研究。分析表明,图表摘要训练数据集的目标端常包含额外信息,从而导致幻觉。我们提出一种基于自然语言推理(NLI)的方法对训练数据进行预处理,并通过人工评估证明该方法显著减少了幻觉。我们还发现,缩短输入序列中的长距离依赖并添加图表相关信息(如标题和图例)能够提升整体性能。

英文摘要:

Hallucinations in text generation occur when the system produces text that is not grounded in the input. In this work, we tackle the problem of hallucinations in neural chart summarization. Our analysis shows that the target side of chart summarization training datasets often contains additional information, leading to hallucinations. We propose a natural language inference (NLI) based method to preprocess the training data and show through human evaluation that our method significantly reduces hallucinations. We also found that shortening long-distance dependencies in the input sequence and adding chart-related information like title and legends improves the overall performance.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑