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 2212.07249cs.CLcs.LG
收藏

APOLLO:一种面向长文本数值推理的优化训练方法

APOLLO: An Optimized Training Approach for Long-form Numerical Reasoning

School of Informatics, Xiamen University(厦门大学信息学院);IDEA Research(IDEA研究院);College of Computer Science, Inner Mongolia University(内蒙古大学计算机学院);Microsoft Research Asia(微软亚洲研究院)

Jiashuo Sun, Hang Zhang, Chen Lin, Xiangdong Su, Yeyun Gong, Jian Guo

AI总结:

针对现有长文本数值推理检索-生成框架存在的事实区分不足、程序一致性缺失问题,提出APOLLO方法,通过数字感知负采样、一致性强化学习等策略,在FinQA和ConvFinQA基准上取得新的最优结果。

摘要:

财务分析中的长文本数值推理旨在生成推理程序,以计算给定问题的正确答案。现有研究采用检索器-生成器框架,由检索器从长文本中选取关键事实,生成器基于检索到的事实生成推理程序。但这类方法对所有事实一视同仁,未考虑含数字与不含数字的事实贡献存在差异;同时监督训练中忽略了程序一致性,导致训练准确率与多样性偏低。为解决上述问题,我们提出APOLLO以改进长文本数值推理框架。在检索器层面,采用数字感知负采样策略,使检索器对关键数值事实更具区分度;在生成器层面,基于程序执行结果的一致性,设计了基于一致性的强化学习与目标程序增强策略。在FinQA和ConvFinQA排行榜上的实验结果验证了所提方法的有效性,取得了新的最优性能。

英文摘要:

Long-form numerical reasoning in financial analysis aims to generate a reasoning program to calculate the correct answer for a given question. Previous work followed a retriever-generator framework, where the retriever selects key facts from a long-form document, and the generator generates a reasoning program based on retrieved facts. However, they treated all facts equally without considering the different contributions of facts with and without numbers. Meanwhile, the program consistency were ignored under supervised training, resulting in lower training accuracy and diversity. To solve these problems, we proposed APOLLO to improve the long-form numerical reasoning framework. For the retriever, we adopt a number-aware negative sampling strategy to enable the retriever to be more discriminative on key numerical facts. For the generator, we design consistency-based reinforcement learning and target program augmentation strategy based on the consistency of program execution results. Experimental results on the FinQA and ConvFinQA leaderboard verify the effectiveness of our proposed method, achieving the new state-of-the-art.

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。

↑