论文专业
graded correctness
分级正确性：一种评估理念，把输出按连续刻度打分（完全正确 / 部分正确 / 错误），承认同一任务下多个答案可能不同程度的有效。常用在 LLM 与 NL2Vis 评估中。
English Definition
"An evaluation notion in which outputs are scored along a continuum (e.g., fully correct, partially correct, incorrect) rather than as a binary right/wrong, acknowledging that multiple answers may be valid to varying degrees."
用法说明 · 针对中文母语者
和 binary correctness / multiple valid answers 一起讨论。Lexara 的核心立场之一就是支持 graded correctness。
真实用例 · 2 条
"existing visualization-specific metrics (...) rarely accommodate graded correctness and are difficult to interpret"
论文Pipeline extracted — review needed·现有可视化指标的局限
"practitioners emphasize the need for flexible, multi-granular evaluation that accommodates graded correctness and multiple valid answers"
论文Pipeline extracted — review needed·实践者需求
由 pipeline 自动采集，待人工 review