推文中药物名称的命名实体识别式自动提取
计算与语言
2021-12-01 v1
摘要
社交媒体帖子包含关于医疗状况与健康相关行为的潜在有价值信息。Biocreative VII任务3聚焦于通过识别推文中药物与膳食补充剂的提及来挖掘此类信息。我们将该任务视为微调多个BERT风格语言模型进行令牌级分类，并将它们集成为整体以生成最终预测。我们最优系统由五个Megatron-BERT-345M模型组成，在未见测试数据上取得了0.764的严格F1分数。
引用
@article{arxiv.2111.15641,
title = {Automatic Extraction of Medication Names in Tweets as Named Entity Recognition},
author = {Carol Anderson and Bo Liu and Anas Abidin and Hoo-Chang Shin and Virginia Adams},
journal= {arXiv preprint arXiv:2111.15641},
year = {2021}
}
备注
Submission to the BioCreative VII challenge - Track-3