COS960

COS960是由清华大学和北京航空航天大学联合创建的中文词相似度数据集，包含960对由两个语素组成的词对。数据集通过使用著名的语言知识库HowNet进行词选择，并根据词性标签分为名词、动词和形容词三类。创建过程中，每对词由15位母语为中文的大学生进行相似度评分，评分范围为0-4。COS960数据集旨在解决中文多词表达（MWEs）的相似度问题，为自然语言处理领域提供了一个有价值的资源。COS960 is a Chinese word similarity dataset jointly created by Tsinghua University and Beihang University. It contains 960 word pairs, each consisting of two morphemes. The dataset selects its lexical items using the well-known linguistic knowledge base HowNet, and is categorized into three classes: nouns, verbs and adjectives based on their part-of-speech tags. During the dataset construction, each word pair was rated for similarity by 15 native Chinese-speaking college students, with the scoring range spanning from 0 to 4. The COS960 dataset aims to address the similarity assessment problem of Chinese multi-word expressions (MWEs), providing a valuable resource for the natural language processing field.

搜集汇总

背景与挑战

背景概述

COS960是一个中文词对相似度数据集，包含960个词对，每个词对由15名母语者标注相似度分数，反映真实相似性。数据集按词性分为名词、动词和形容词三组，用于测试词嵌入模型，并提供了学术引用和计算工具。