thunlp / PL-Marker

Source code for "Packed Levitated Marker for Entity and Relation Extraction"
MIT License
260 stars 35 forks source link

语料选择中文数据集,预训练模型bert-base-Chineseese,效果太差,有哪些步骤需要修改的吗? #58

Closed thankslife closed 1 year ago

thankslife commented 1 year ago

我把语料换成中文数据集CMEIE,预训练模型bert-base-chinese,为什么效果这么差,pure F值还能达到0.6呢, 下面是pl-marker ner结果 {"dev_bestf1": 0.08358639461309555, "f1": 0.08358639461309555, "f1overlap": 0.003212469772290498, "precision": 0.04741821366428135, "recall": 0.3523116438356164}

ToddLynn commented 1 year ago

我把语料换成中文数据集CMEIE,预训练模型bert-base-Chineseese,为什么效果这么差,pure F值还能达到0.6呢, 下面是pl-marker ner结果 {“dev_bestf1”: 0.08358639461309555, “f1”: 0.08358639461309555, “f1overlap”: 0.003212469772290498, “precision”: 0.04741821366428135, “recall”: 0.3523116438356164}

您也有尝试将语料换成CMeIE的数据进行测试么?最新的效果有提升么请问

ToddLynn commented 1 year ago

我把语料换成中文数据集CMEIE,预训练模型bert-base-Chineseese,为什么效果这么差,pure F值还能达到0.6呢, 下面是pl-marker ner结果 {“dev_bestf1”: 0.08358639461309555, “f1”: 0.08358639461309555, “f1overlap”: 0.003212469772290498, “precision”: 0.04741821366428135, “recall”: 0.3523116438356164}

810304762

YeDeming commented 1 year ago

这个代码没有在中文上进行过实验