在Github上搜索整理了一波關(guān)于醫(yī)療NLP的數(shù)據(jù)集：

1

中文評(píng)測(cè)數(shù)據(jù)集

1. Yidu-S4K：醫(yī)渡云結(jié)構(gòu)化4K數(shù)據(jù)集

2.瑞金醫(yī)院糖尿病數(shù)據(jù)集

3.Yidu-N7K：醫(yī)渡云標(biāo)準(zhǔn)化7K數(shù)據(jù)集

4.中文醫(yī)學(xué)問(wèn)答數(shù)據(jù)集

5.平安醫(yī)療科技疾病問(wèn)答遷移學(xué)習(xí)比賽

6.天池新冠肺炎問(wèn)句匹配比賽

7.中文醫(yī)患問(wèn)答對(duì)話數(shù)據(jù)

8.中文醫(yī)學(xué)問(wèn)答數(shù)據(jù)

9.CHIP2020各項(xiàng)評(píng)測(cè)已開(kāi)放

10.醫(yī)學(xué)數(shù)據(jù)挖掘與算法評(píng)測(cè)大賽

中文醫(yī)學(xué)知識(shí)圖譜

CMeKG

英文數(shù)據(jù)集

PubMedQA: A Dataset for Biomedical Research Question Answering

中文醫(yī)療領(lǐng)域語(yǔ)料

醫(yī)學(xué)教材 培訓(xùn)考試

哈工大《大詞林》開(kāi)放75萬(wàn)核心實(shí)體詞及相關(guān)概念、關(guān)系列表（包含中藥/醫(yī)院/生物 類別）

醫(yī)學(xué)embedding

開(kāi)源英文醫(yī)學(xué)embedding

鏈接：https://github.com/lrs1353281004/Chinese_medical_NLP

2

醫(yī)療行業(yè)專業(yè)詞匯語(yǔ)料

鏈接：https://github.com/xtea/chinese_medical_words

3

中文醫(yī)學(xué)NLP公開(kāi)資源整理：術(shù)語(yǔ)集/語(yǔ)料庫(kù)/詞向量/預(yù)訓(xùn)練模型/知識(shí)圖譜/命名實(shí)體識(shí)別/QA/信息抽取/etc

術(shù)語(yǔ)集/語(yǔ)料庫(kù)

medical-news中文醫(yī)學(xué)新聞爬蟲(chóng)

medical-books中文LaTex開(kāi)源醫(yī)學(xué)書(shū)籍

THUOCL清華大學(xué)thunlp組醫(yī)學(xué)詞匯

ICD-10-CNICD-10中文對(duì)應(yīng)

OMAHA七巧板醫(yī)學(xué)術(shù)語(yǔ)集樣例數(shù)據(jù)

中文糖尿病標(biāo)注數(shù)據(jù)集包含實(shí)體標(biāo)注和關(guān)系標(biāo)注

詞向量/預(yù)訓(xùn)練模型

ChineseEHRBert 中文電子病歷預(yù)訓(xùn)練Bert；用Bert測(cè)試命名實(shí)體識(shí)別，問(wèn)答模型，關(guān)系提取任務(wù)

分詞

PKUSEGPKUSEG分詞工具，模型支持選擇醫(yī)學(xué)

知識(shí)圖譜 / 關(guān)系提取

cMeKGChinese Medical Knowledge Graph

瑞金醫(yī)院人工智能輔助構(gòu)建知識(shí)圖譜大賽糖尿病相關(guān)的學(xué)術(shù)論文以及糖尿病臨床指南的實(shí)體標(biāo)注和抽取實(shí)體關(guān)系任務(wù)

OMAHA知識(shí)圖譜（藥品適應(yīng)癥）開(kāi)放醫(yī)療與健康聯(lián)盟（Open Medical and Healthcare Alliance，OMAHA）構(gòu)建的藥品與藥品適應(yīng)證的知識(shí)圖譜數(shù)據(jù)

醫(yī)療知識(shí)圖譜數(shù)據(jù)醫(yī)療知識(shí)圖譜數(shù)據(jù)（ownthink）

病人事件圖譜數(shù)據(jù)集病人事件圖譜是一種新的基于RDF的醫(yī)療觀察性數(shù)據(jù)表示模型，可以清晰地表示臨床檢查、診斷、治療等多種事件類型以及事件的時(shí)序關(guān)系。使用三家上海三甲醫(yī)院的電子病歷數(shù)據(jù)，構(gòu)建了包括3個(gè)專科、173395個(gè)醫(yī)療事件、501335個(gè)事件時(shí)序關(guān)系以及與5313個(gè)知識(shí)庫(kù)概念鏈接的醫(yī)療數(shù)據(jù)集。

中文癥狀庫(kù)這是一個(gè)包含癥狀實(shí)體和癥狀相關(guān)三元組的數(shù)據(jù)集。中文癥狀庫(kù)的數(shù)據(jù)來(lái)自8個(gè)主流的健康咨詢網(wǎng)站、3個(gè)中文百科網(wǎng)站和電子病歷。它還包含了中文癥狀與UMLS中概念的鏈接結(jié)果。

中醫(yī)醫(yī)案知識(shí)圖譜從醫(yī)案中抽取臨床知識(shí)構(gòu)建知識(shí)圖譜，幫助用戶了解中醫(yī)特色療法，以及疾病（如“慢性胃炎”）的臨床表現(xiàn)、相關(guān)療法、相關(guān)養(yǎng)生保健方法等

herbnet 面向中藥研究，根據(jù)中藥領(lǐng)域模型的特點(diǎn)，構(gòu)建了一個(gè)包括中醫(yī)疾病，方劑，中藥， 中藥化學(xué)成分，藥理作用，中藥實(shí)驗(yàn)，化學(xué)實(shí)驗(yàn)方法在內(nèi)的中藥本體。進(jìn)而，基于本體實(shí)現(xiàn)了一系列數(shù)據(jù)庫(kù)的集成，從而構(gòu)建了一個(gè)中藥知識(shí)圖譜。

CHIP2020中文醫(yī)學(xué)文本實(shí)體關(guān)系抽取

命名實(shí)體識(shí)別

CCKS2017面向中文電子病歷的醫(yī)療實(shí)體識(shí)別及屬性抽取數(shù)據(jù)集

CCKS2018面向中文電子病歷的醫(yī)療實(shí)體識(shí)別及屬性抽取數(shù)據(jù)集

CCKS2019數(shù)據(jù)下載面向中文電子病歷的醫(yī)療實(shí)體識(shí)別及屬性抽取數(shù)據(jù)集

CHIP2020中文醫(yī)學(xué)文本命名實(shí)體識(shí)別

CHIP2020中藥說(shuō)明書(shū)實(shí)體識(shí)別

QA

CCIR2019CCIR 2019 基于電子病歷的數(shù)據(jù)查詢類問(wèn)答

cMedQA中文醫(yī)學(xué)QA數(shù)據(jù)集

cMedQA2中文醫(yī)學(xué)QA數(shù)據(jù)集

CMID中文醫(yī)學(xué)QA意圖理解數(shù)據(jù)集

KGQA基于醫(yī)藥知識(shí)圖譜的智能問(wèn)答系統(tǒng)

chatbot-base-on-Knowledge-Graph使用深度學(xué)習(xí)方法解析問(wèn)題 知識(shí)圖譜存儲(chǔ) 查詢知識(shí)點(diǎn) 基于醫(yī)療垂直領(lǐng)域的對(duì)話系統(tǒng)

中文醫(yī)療對(duì)話數(shù)據(jù)集Chinese medical dialogue data 中文醫(yī)療對(duì)話數(shù)據(jù)集

webMedQAwebMedQA

MedDialogThe MedDialog dataset contains conversations (in Chinese) between doctors and patients. It has 1.1 million dialogues and 4 million utterances.

CHIP2020中醫(yī)文獻(xiàn)問(wèn)題生成

術(shù)語(yǔ)標(biāo)準(zhǔn)化

CHIP2019臨床術(shù)語(yǔ)標(biāo)準(zhǔn)化任務(wù):醫(yī)渡云標(biāo)準(zhǔn)化7K數(shù)據(jù)集

CHIP2020臨床術(shù)語(yǔ)標(biāo)準(zhǔn)化任務(wù)

相似句對(duì)判斷

“公益AI之星”挑戰(zhàn)賽-新冠疫情相似句對(duì)判定大賽比賽整理近萬(wàn)條真實(shí)語(yǔ)境下疫情相關(guān)的肺炎、支原體肺炎、支氣管炎、上呼吸道感染、肺結(jié)核、哮喘、胸膜炎、肺氣腫、感冒、咳血等患者提問(wèn)句對(duì)，要求選手通過(guò)自然語(yǔ)言處理技術(shù)識(shí)別相似的患者問(wèn)題。

其他

CHIP2018針對(duì)中文的真實(shí)患者健康咨詢語(yǔ)料，進(jìn)行問(wèn)句意圖匹配

CHIP2019平安醫(yī)療科技疾病問(wèn)答遷移學(xué)習(xí)比賽

鏈接：https://github.com/GanjinZero/awesome_Chinese_medical_NLP

4

Data_數(shù)據(jù)中有6個(gè)文件夾分別是:

94596個(gè)問(wèn)答對(duì)220606個(gè)問(wèn)答對(duì)183751個(gè)問(wèn)答對(duì)75553個(gè)問(wèn)答對(duì)101602個(gè)問(wèn)答對(duì)115991個(gè)問(wèn)答對(duì) 總計(jì) 792099個(gè)問(wèn)答對(duì)

每個(gè)文件夾下有一個(gè)csv文件，其中的數(shù)據(jù)格式為:

鏈接：https://github.com/Toyhom/Chinese-medical-dialogue-data

5

This dataset is used for Chinese medical QA intent understanding task.

Dataset format:

All the data is stored in a JSON file. There are 5 fields in the file. An example as follows:

{ "originalText": "間質(zhì)性肺炎的癥狀?", "entities": [{"label_type": "疾病和診斷", "start_pos": 0, "end_pos": 5}], "seg_result": ["間質(zhì)性肺炎", "的", "癥狀", "?"], "label_4class": ["病癥"], "label_36class": ["臨床表現(xiàn)"] }

鏈接：https://github.com/liutongyang/CMID

6

This is the dataset for Chinese community medical question answering. The dataset is in version 1.0 and is available for non-commercial research. We will update and expand the database from time to time. In order to protect the privacy, the data is anonymized and no personal information is included.

鏈接：https://github.com/zhangsheng93/cMedQA

7

COVID19 Language Resources: Datasets

鏈接：https://github.com/lwgkzl/Covid19-NLP

8

Datasets

BioCreative V chemical-disease relation (CDR) corpus (in short, BC5CDR corpus) (13, 14, 16, 34): It consists of 1,500 PubMed articles with 4,409 annotated chemicals, 5,818 diseases, and 3,116 chemical-disease interactions. The relation task data is publicly available through BioCreative V athttps://biocreative.bioinformatics.udel.edu/resources/corpora/biocreative-v-cdr-corpus/.

Traditional Chinese medicine (TCM) literature corpus (in short, TCM corpus) (32): The abstracts of all 106,150 papers published in the 114 most popular Chinese TCM journals between 2011 to 2016 are collected. 3024 herbs, 4957 formulae, 1126 syndromes, and 1650 diseases are found. 5 types of relations are annotated. The entire dataset is available online athttp://arnetminer.org/TCMRelExtr.

The 2012 informatics for integrating biology and the bedside (i2b2) project temporal relations challenge corpus (in short, i2b2 temporal corpus) (29, 30): It contains 310 de-identified discharge summaries of more than 178,000 tokens, with annotations of clinically significant events, temporal expressions and temporal relations in clinical narratives. On average, each discharge summary in the corpus contains 86.6 events, 12.4 temporal expressions, and 176 raw temporal relations. In this corpus, 8 kinds of temporal relations between events and temporal expressions are defined: BEFORE, AFTER, SIMULTANEOUS, OVERLAP, BEGUN_BY, ENDED_BY, DURING, BEFORE_OVERLAP. The entire annotations are available athttp://i2b2.org/NLP/DataSets.

鏈接：https://github.com/chentao1999/MedicalRelationExtraction