命名实体识别实验

[fseasy]

• [ ] 模型一
• [ ] 模型二
• [ ] 模型三
• [ ] 模型四
• [ ] 模型五

[fseasy]

模型一

基础的完全基于分类的模型

待补充

[fseasy]

模型二

实验设置

1. word embedding dim : 50 , postag embedding dim : 5 , ner embedding dim : 5
2. max epoch 5 , devel freq : 6000 instances
3. disable dropout

实验结果

dataset	pre-tag F1
pku-train	98.02%
pku-holdout	92.94%
pku-test	93.08%

[fseasy]

模型三

实验设置

1. gigawords : word2vec training with negative-samples mode and skip-gram mothod , dimension 50
2. sogou-news : same as gigawords
3. the others is same as 实验二 setting

实验结果

datastet	dc-giga-skipgram F1	dc-sogou-skipgram F1
pku-train	97.62%	99.64%
pku-holdout	94.44%	94.26%
pku-test	94.07%	94.33%

[fseasy]

模型四

实验设置

1. dropout rate 0.1
2. the others is same as 实验二 setting

实验结果

dataset	crf-dr0.1 F1
pku-train	97.80%
pku-holdout	93.34%
pku-test	94.07%

• dr0.x 表示dropout rate 0.x

[fseasy]

模型五

实验设置

1. dropout rate 0.1
2. the others is the same as 实验三 setting

实验结果

dataset	giga-dr0.1	sogou-dr0.1
pku-train	98.69%	98.06%
pku-holdout	94.37%	94.21%
pku-test	94.47%	94.71%