Out-of-Scope Intent Classification Dataset 范围外意图分类数据集

大多数受监督的机器学习任务都采用一组定义明确的目标标签集的数据集。但是，当训练有素的模型与现实世界相遇时会发生什么情况呢？此数据集提供了评估"范围外"输入的意向分类模型的方法。

"范围外"输入是不属于"范围内"目标标签集的输入。您可能听说过其他指超范围的方式，包括"域外"或"外分配"。

Most supervised machine learning tasks employ datasets with a well-defined set of target labels. However, what happens when a trained model encounters real-world inputs? This dataset offers a framework for evaluating intent classification models on out-of-scope inputs. "Out-of-scope inputs" are defined as inputs that do not fall within the set of in-scope target labels. You may have encountered other terms used to refer to out-of-scope cases, including "out-of-distribution" or "out-of-assignment".

搜集汇总

背景与挑战

背景概述

该数据集用于评估意图分类模型处理'范围外'输入的能力，即不属于预定义目标标签集的文本。它包含'范围内'数据（150个意图类别）和'范围外'数据（一个oos意图类别），支持模型在真实场景中的泛化性能测试。