Recent studies have found that summaries generated by large language models (LLMs) are favored by human annotators over the original reference summaries in commonly used summarization datasets. Therefore, we investigate a new learning paradigm of text summarization models that considers the LLMs as the reference or the gold-standard oracle on commonly used summarization datasets such as the CNN/DailyMail dataset. To examine the standard practices that are aligned with the new learning setting, we propose a novel training method that is based on contrastive learning with LLMs as a summarization quality evaluator. For this reward-based training method, we investigate two different methods of utilizing LLMs for summary quality evaluation, namely GPTScore and GPTRank. Our experiments on the CNN/DailyMail dataset demonstrate that smaller summarization models trained by our proposed method can achieve performance equal to or surpass that of the reference LLMs, as evaluated by the LLMs themselves. This underscores the efficacy of our proposed paradigm in enhancing model performance over the standard maximum likelihood estimation (MLE) training method, and its efficiency since it only requires a small budget to access the LLMs. We release the training scripts, model outputs, and LLM-based evaluation results to facilitate future studies.

翻译：近期研究发现，在常用摘要数据集中，人类标注者更倾向选择大语言模型生成的摘要而非原始参考摘要。为此，我们提出一种新的文本摘要模型训练范式，将大语言模型作为常用摘要数据集（如CNN/DailyMail数据集）的参考标准或黄金标准。为探究与新学习设定相适配的标准实践，我们提出基于对比学习的创新训练方法，该方法以大语言模型作为摘要质量评估器。针对这种基于奖励的训练方法，我们研究了利用大语言模型进行摘要质量评估的两种不同方式：GPTScore与GPTRank。在CNN/DailyMail数据集上的实验表明，采用我们提出的方法训练的小型摘要模型，其性能可达到甚至超越参考大语言模型（以大语言模型自身评估为准）。这证实了我们的新范式相较于标准最大似然估计训练方法在提升模型性能方面的有效性，同时其效率优势明显——仅需少量预算即可调用大语言模型。为便于后续研究，我们开源了训练脚本、模型输出及基于大语言模型的评估结果。