AW-OIE (All Words OpenIE)

All Words Open IE (AW-OIE) 是一个从问答含义表示 (QAMR) 数据集派生的开放信息提取数据集。我们提供了支持开放信息提取 (Open IE) 的监督学习方法的数据和方法。该方法的核心是将 Open IE 作为序列标记问题的新公式，解决诸如为谓词编码多个提取等挑战。我们还开发了一个 bi?LSTM 转换器，扩展了最近的深度语义角色标签模型，以提取 Open IE 元组并提供置信度分数以调整其精确召回权衡。此外，我们还展示了最近发布的问答含义表示数据集可以自动转换为开放式 IE 语料库，这显着增加了可用训练数据的数量。我们的监督模型已公开发布,1 在基准数据集上优于 Open IE 中的最新技术。_x000D_All Words Open IE (AW-OIE) is an open information extraction dataset derived from the Question Answering Meaning Representation (QAMR) dataset. We provide datasets and methodologies for supervised learning approaches that support open information extraction (Open IE). The core of this approach is a novel formulation of Open IE as a sequence tagging problem, which addresses challenges such as encoding multiple extractions for a single predicate. We also developed a bi?LSTM Transformer that extends recent deep semantic role labeling models to extract Open IE tuples and provide confidence scores to adjust their precision-recall trade-off. Furthermore, we demonstrate that the recently released Question Answering Meaning Representation dataset can be automatically converted into an open IE corpus, which significantly increases the volume of available training data. Our supervised model has been publicly released,¹ and outperforms the state-of-the-art in Open IE on benchmark datasets.

搜集汇总

背景与挑战

背景概述

AW-OIE是一个从问答含义表示数据集派生的开放信息提取数据集，支持将开放信息提取作为序列标记问题的监督学习方法，并采用bi-LSTM转换器模型提取元组。该数据集通过自动转换增加了训练数据量，其模型在基准测试中优于现有技术。