数据集:

• tner/tweetner7 评估指标:
• F1值
• 精确率
• 召回率 模型索引:
• 名称: tner/twitter-roberta-base-dec2021-tweetner7-random 结果:
  • 任务: 名称: 标记分类 类型: token-classification 数据集: 名称: tner/tweetner7 类型: tner/tweetner7 参数: tner/tweetner7 指标:
    • 名称: F1值 (test_2021) 类型: f1 值: 0.6321284238886395
    • 名称: 精确率 (test_2021) 类型: precision 值: 0.6142015706806283
    • 名称: 召回率 (test_2021) 类型: recall 值: 0.6511332099907493
    • 名称: 宏观F1值 (test_2021) 类型: f1_macro 值: 0.583682304736069
    • 名称: 宏观精确率 (test_2021) 类型: precision_macro 值: 0.5654677691354458
    • 名称: 宏观召回率 (test_2021) 类型: recall_macro 值: 0.6047150410746663
    • 名称: 实体跨度F1值 (test_2021) 类型: f1_entity_span 值: 0.7703620544484986
    • 名称: 实体跨度精确率 (test_2020) 类型: precision_entity_span 值: 0.7484729493891797
    • 名称: 实体跨度召回率 (test_2021) 类型: recall_entity_span 值: 0.7935700242858795
    • 名称: F1值 (test_2020) 类型: f1 值: 0.6368775235531628
    • 名称: 精确率 (test_2020) 类型: precision 值: 0.6616331096196868
    • 名称: 召回率 (test_2020) 类型: recall 值: 0.6139076284379865
    • 名称: 宏观F1值 (test_2020) 类型: f1_macro 值: 0.5976605759407211
    • 名称: 宏观精确率 (test_2020) 类型: precision_macro 值: 0.6177069721428509
    • 名称: 宏观召回率 (test_2020) 类型: recall_macro 值: 0.5812570646484104
    • 名称: 实体跨度F1值 (test_2020) 类型: f1_entity_span 值: 0.7542395693135936
    • 名称: 实体跨度精确率 (test_2020) 类型: precision_entity_span 值: 0.7835570469798657
    • 名称: 实体跨度召回率 (test_2020) 类型: recall_entity_span 值: 0.7270368448365335

管道标签: token-classification 小部件:

• 文本: "从{@herbiehancock@}通过{@bluenoterecords@}获取Takin' Off专辑的全模拟经典黑胶版，链接如下: {{URL}}" 示例标题: "命名实体识别示例1"

tner/twitter-roberta-base-dec2021-tweetner7-random

该模型是基于cardiffnlp/twitter-roberta-base-dec2021在tner/tweetner7数据集（train_random分割）上微调的版本。模型微调通过T-NER的超参数搜索完成（详见仓库）。在2021年的测试集上取得了以下结果：

• F1（微观）: 0.6321284238886395
• 精确率（微观）: 0.6142015706806283
• 召回率（微观）: 0.6511332099907493
• F1（宏观）: 0.583682304736069
• 精确率（宏观）: 0.5654677691354458
• 召回率（宏观）: 0.6047150410746663

测试集上各实体类别的F1分数细分如下：

• 公司: 0.5019685039370079
• 创意作品: 0.41401273885350315
• 事件: 0.4564727108705458
• 团体: 0.5892444737710327
• 地点: 0.6486486486486486
• 人物: 0.8268075031870332
• 产品: 0.6486215538847118

F1分数的置信区间通过自助法获得如下：

• F1（微观）:
  • 90%: [0.6245116881258609, 0.6411928894306437]
  • 95%: [0.6221686986039963, 0.642603475030015]
• F1（宏观）:
  • 90%: [0.6245116881258609, 0.6411928894306437]
  • 95%: [0.6221686986039963, 0.642603475030015]

完整评估结果可在NER指标文件和实体跨度指标文件中查看。

使用方式

该模型可通过tner库使用。通过pip安装库：

pip install tner

TweetNER7预处理了推文，将账户名和URL转换为特殊格式（详见数据集页面），因此我们需要相应处理推文后运行模型预测，如下所示：

import re
from urlextract import URLExtract
from tner import TransformersNER

extractor = URLExtract()

def format_tweet(tweet):
    # 屏蔽网页URL
    urls = extractor.find_urls(tweet)
    for url in urls:
        tweet = tweet.replace(url, "{{URL}}")
    # 格式化推特账户
    tweet = re.sub(r"\b(\s*)(@[\S]+)\b", r'\1{\2@}', tweet)
    return tweet


text = "从@herbiehancock通过@bluenoterecords获取`Takin' Off`专辑的全模拟经典黑胶版，链接如下: http://bluenote.lnk.to/AlbumOfTheWeek"
text_format = format_tweet(text)
model = TransformersNER("tner/twitter-roberta-base-dec2021-tweetner7-random")
model.predict([text_format])

也可以通过transformers库使用，但不推荐，因为目前不支持CRF层。

训练超参数

训练期间使用的超参数如下：

• 数据集: ['tner/tweetner7']
• 数据集分割: train_random
• 数据集名称: None
• 本地数据集: None
• 模型: cardiffnlp/twitter-roberta-base-dec2021
• CRF: True
• 最大长度: 128
• 训练轮数: 30
• 批量大小: 32
• 学习率: 0.0001
• 随机种子: 0
• 梯度累积步数: 1
• 权重衰减: 1e-07
• 学习率预热步比例: 0.15
• 最大梯度范数: 1

完整配置可在微调参数文件中查看。

引用

如果使用该模型，请引用T-NER论文和TweetNER7论文。

• T-NER

@inproceedings{ushio-camacho-collados-2021-ner,
    title = "{T}-{NER}: An All-Round Python Library for Transformer-based Named Entity Recognition",
    author = "Ushio, Asahi  and
      Camacho-Collados, Jose",
    booktitle = "Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations",
    month = apr,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.eacl-demos.7",
    doi = "10.18653/v1/2021.eacl-demos.7",
    pages = "53--62",
    abstract = "Language model (LM) pretraining has led to consistent improvements in many NLP downstream tasks, including named entity recognition (NER). In this paper, we present T-NER (Transformer-based Named Entity Recognition), a Python library for NER LM finetuning. In addition to its practical utility, T-NER facilitates the study and investigation of the cross-domain and cross-lingual generalization ability of LMs finetuned on NER. Our library also provides a web app where users can get model predictions interactively for arbitrary text, which facilitates qualitative model evaluation for non-expert programmers. We show the potential of the library by compiling nine public NER datasets into a unified format and evaluating the cross-domain and cross- lingual performance across the datasets. The results from our initial experiments show that in-domain performance is generally competitive across datasets. However, cross-domain generalization is challenging even with a large pretrained LM, which has nevertheless capacity to learn domain-specific features if fine- tuned on a combined dataset. To facilitate future research, we also release all our LM checkpoints via the Hugging Face model hub.",
}

• TweetNER7

@inproceedings{ushio-etal-2022-tweet,
    title = "{N}amed {E}ntity {R}ecognition in {T}witter: {A} {D}ataset and {A}nalysis on {S}hort-{T}erm {T}emporal {S}hifts",
    author = "Ushio, Asahi  and
        Neves, Leonardo  and
        Silva, Vitor  and
        Barbieri, Francesco. and
        Camacho-Collados, Jose",
    booktitle = "The 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing",
    month = nov,
    year = "2022",
    address = "Online",
    publisher = "Association for Computational Linguistics",
}