用真实新闻对话对微调LLM，提升电视新闻问答的准确性和可验证性。
News Reporter: A Multi-lingual LLM Framework for Broadcast T.V News
- 基于美国多家新闻频道真实录音提取问答对，训练专用LLM。
- 在多个基准测试中超越同规模基础模型，准确率显著提升。
- 融合RAG技术，使回答可追溯到原始新闻片段，适合新闻机构使用。
大型语言模型（LLMs）因能对多样化问题生成连贯回复，已成为众多对话聊天机器人的核心工具。然而，用于训练这些模型的数据集通常混合了通用与合成样本，缺乏对电视新闻类问题的准确性与可验证性支持。本文收集并公开了来自美国多家新闻频道播音录音的大量问答对，利用这些数据对现成的LLM进行微调。微调后的模型在多个开源LLM基准测试中表现优于同规模基线模型。此外，我们提出并集成了一个RAG方法，以增强回答的上下文关联性，并将答案指向可验证的原始新闻录音，确保信息来源可信。
原文摘要 · Abstract (English)
Large Language Models (LLMs) have fast become an essential tools to many conversational chatbots due to their ability to provide coherent answers for varied queries. Datasets used to train these LLMs are often a mix of generic and synthetic samples, thus lacking the verification needed to provide correct and verifiable answers for T.V. News. We collect and share a large collection of QA pairs extracted from transcripts of news recordings from various news-channels across the United States. Resultant QA pairs are then used to fine-tune an off-the-shelf LLM model. Our model surpasses base models of similar size on several open LLM benchmarks. We further integrate and propose a RAG method to improve contextualization of our answers and also point it to a verifiable news recording.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。