ties76587578/conll2003
收藏资源简介:
CoNLL-2003共享任务涉及语言无关的命名实体识别,专注于四种类型的命名实体:人物、地点、组织以及不属于前三类的其他实体。该数据集文件包含四列,由单个空格分隔,每个单词单独一行,每个句子后有空行。每行的第一项是单词,第二项是词性标注(POS),第三项是句法组块标注,第四项是命名实体标注。组块标注和命名实体标注采用I-TYPE格式,表示单词位于TYPE类型的短语内部。只有当两个相同类型的短语紧邻时,第二个短语的第一个单词会使用B-TYPE标签以表示新短语的开始。标签为O的单词不属于任何短语。注意,该数据集使用IOB2标注方案,而原始数据集使用IOB1。数据集基于路透社新闻语料库,包含训练集(14041条)、验证集(3250条)和测试集(3453条)。
CoNLL-2003 shared task focuses on language-independent named entity recognition, targeting four types of named entities: persons, locations, organizations, and miscellaneous entities that do not belong to the first three categories. The dataset files contain four columns separated by single spaces, with each word on its own line and a blank line after each sentence. The first item on each line is the word, the second is the part-of-speech (POS) tag, the third is the syntactic chunk tag, and the fourth is the named entity tag. Both chunk tags and named entity tags adopt the I-TYPE format, indicating that the word is inside a phrase of TYPE. Only when two phrases of the same type are adjacent will the first word of the second phrase use the B-TYPE label to indicate the start of a new phrase. Words labeled O do not belong to any phrase. Please note that this dataset uses the IOB2 annotation scheme, while the original dataset uses IOB1. The dataset is based on the Reuters news corpus, and includes the training set (14041 instances), validation set (3250 instances), and test set (3453 instances).




