arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2408.10490cs.CLcs.IR
收藏

基于计划的检索用于接地文本生成的分析

Analysis of Plan-based Retrieval for Grounded Text Generation

Google Research(谷歌研究院);Google DeepMind(谷歌DeepMind);University of Southern California(南加州大学)

Ameya Godbole, Nicholas Monath, Seungyeon Kim, Ankit Singh Rawat, Andrew McCallum, Manzil Zaheer

AI总结:

本文利用指令微调大语言模型的规划能力引导检索,以降低长文本生成中的幻觉,通过提高相关事实覆盖率,生成更具信息量且归因率更高的响应。

摘要:

在文本生成中,幻觉指的是生成看似连贯但与已有知识相矛盾的文本。一个引人注目的假设是,当语言模型被赋予一个超出其参数知识范围(由于稀有性、新近性、领域等)的生成任务时,就会发生幻觉。解决这一局限性的常见策略是向语言模型注入检索机制,为模型提供与任务相关的知识。在本文中,我们利用指令微调的大语言模型的规划能力,并分析规划如何用于引导检索,以进一步降低幻觉的发生频率。我们在长文本生成任务上对我们的方法进行了多种变体的实证评估。通过提高相关事实的覆盖率,基于计划的检索和生成可以产生更具信息量的响应,同时提供更高的源文档归因率。

英文摘要:

In text generation, hallucinations refer to the generation of seemingly coherent text that contradicts established knowledge. One compelling hypothesis is that hallucinations occur when a language model is given a generation task outside its parametric knowledge (due to rarity, recency, domain, etc.). A common strategy to address this limitation is to infuse the language models with retrieval mechanisms, providing the model with relevant knowledge for the task. In this paper, we leverage the planning capabilities of instruction-tuned LLMs and analyze how planning can be used to guide retrieval to further reduce the frequency of hallucinations. We empirically evaluate several variations of our proposed approach on long-form text generation tasks. By improving the coverage of relevant facts, plan-guided retrieval and generation can produce more informative responses while providing a higher rate of attribution to source documents.

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。

↑