本文将讨论推土机距离 Earth Mover’s Distance (EMD),和欧式距离一样,它们都是一种距离度量的定义、可以用来测量某两个分布之间的距离。本文记录推土机距离相关内容。 推土机距离 如果我们将分布想象为两个有一定存土量的土堆,每个土堆维度为 N,那么 EMD 就是将一个
3.1 Earth Mover Distance (EMD,推土机距离) The EMD between two distributions is proportional to the minimum amount ofworkrequired to convert one distribution into the other. 1 unit ofworkis the amount of work necessary to move one unit of weight by one unit of distance. Intuitively, the weight...
文本相似度计算的演变，从最基本的 one-hot 编码到更复杂的词嵌入与预训练模型，可分为三个阶段。本文聚焦于文本相似度度量的第二种方法，Earth Mover Distance(EMD)与Word Mover Distance(WMD)。EMD，即推土机距离，是衡量两个分布之间的相似度。其直观解释为将一个分布转换为另一个所需最小工作量，...
3.1 Earth Mover Distance (EMD,推土机距离) The EMD between two distributions is proportional to the minimum amount ofworkrequired to convert one distribution into the other. 1 unit ofworkis the amount of work necessary to move one unit of weight by one unit of distance. Intuitively, the weight...