arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2204.04413cs.CLcs.AI
收藏

PSP:用于Few-Shot抽象摘要的预训练软模板

PSP: Pre-trained Soft Prompts for Few-Shot Abstractive Summarization

Beijing Institute of Technology(北京理工大学)

Xiaochen Liu, Yang Gao, Yu Bai, Jiawei Li, Yinan Hu, Heyan Huang, Boxing Chen

摘要:

Few-Shot摘要摘要是自然语言生成中的一个具有挑战性的课题。为了支持它,我们设计了一个新的软提示架构,加上一个快速的预训练加微调范式,这是有效的,只调整非常轻的参数。软提示包括跨编码器和解码器的连续输入嵌入,以适应生成模型的结构。重要的是,一个新的内部提示放置在文本中引入捕获文档级的信息。其目的是将注意力放在理解文档上,从而更好地提示模型生成与文档相关的内容。摘要过程的第一步是使用自监督伪数据进行快速预训练。这将教授模型基本的汇总功能。然后使用Few-Shot示例对模型进行微调。在CNN/DailyMail和XSum数据集上的实验结果表明,我们的方法,只有0.1%的参数,优于全模型调整,其中所有的模型参数进行了调整。它还大幅超越了提示微调,并以3%的参数提供了与Prefix-Tuning竞争的结果。

英文摘要:

Few-shot abstractive summarization has become a challenging task in natural language generation. To support it, we designed a novel soft prompts architecture coupled with a prompt pre-training plus fine-tuning paradigm that is effective and tunes only extremely light parameters. The soft prompts include continuous input embeddings across an encoder and a decoder to fit the structure of the generation models. Importantly, a novel inner-prompt placed in the text is introduced to capture document-level information. The aim is to devote attention to understanding the document that better prompts the model to generate document-related content. The first step in the summarization procedure is to conduct prompt pre-training with self-supervised pseudo-data. This teaches the model basic summarizing capabilities. The model is then fine-tuned with few-shot examples. Experimental results on the CNN/DailyMail and XSum datasets show that our method, with only 0.1% of the parameters, outperforms full-model tuning where all model parameters are tuned. It also surpasses Prompt Tuning by a large margin and delivers competitive results against Prefix-Tuning with 3% of the parameters.

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。

↑