ReClor

ReClor数据集来源于标准化的研究生入学考试，旨在提升阅读理解的复杂性，并为模型的逻辑推理能力引入新的挑战。为了降低模型通过利用数据中的固有偏差而获得高分，而没有真正理解文本的风险，数据集的偏差已被创建者划分为简单子集和困难子集。

The ReClor dataset is sourced from standardized graduate admission examinations, designed to enhance the complexity of reading comprehension tasks and introduce novel challenges for the logical reasoning capabilities of models. To mitigate the risk that models may achieve high scores by exploiting inherent biases within the dataset without genuinely comprehending the text, the dataset's creators have partitioned it into simple and difficult subsets.

搜集汇总

背景与挑战

背景概述

ReClor是一个基于标准化考试逻辑推理题目的阅读理解数据集，旨在提升模型的复杂推理能力，仅限非商业研究使用。