Official repository of paper "Zero-Shot Information Extraction via Chatting with ChatGPT". Please star, watch, and fork our repo for the active updates!

Zero-shot information extraction (IE) aims to build IE systems from the unannotated text. It is challenging due to involving little human intervention. Challenging but worthwhile, zero-shot IE reduces the time and effort that data labeling takes. Recent efforts on large language models (LLMs, e.g., GPT3, ChatGPT) show promising performance on zero-shot settings, thus inspiring us to explore prompt-based methods. In this work, we ask whether strong IE models can be constructed by directly prompting LLMs. Specifically, we transform the zero-shot IE task into a multi-turn question-answering problem with a two-stage framework (ChatIE). With the power of ChatGPT, we extensively evaluate our framework on three IE tasks: entityrelation triple extract, named entity recognition, and event extraction. Empirical results on six datasets across two languages show that ChatIE achieves impressive performance and even surpasses some full-shot models on several datasets (e.g., NYT11-HRL). We believe that our work could shed light on building IE models with limited resources.

零れい样本信しん息いき抽取（Information Extraction，IE）旨むね在ざい从无标注文ちゅうもん本中ほんなか建立こんりゅうIE系けい统，因いん为很少しょう涉わたる及人为干预，该问题非常ひじょう具有ぐゆう挑战性せい。但ただし零れい样本IE不ふ再さい需要じゅよう标注数すう据すえ时耗费的时间和人わじん力りょく，因いん此十じゅう分ふん重要じゅうよう。近来きんらい的てき大だい规模语言模型もけい（例れい如GPT-3，Chat GPT）在ざい零れい样本设置下か取得しゅとく了りょう很好的てき表ひょう现，这启发我们探索たんさく基もと于提示ていじ的てき方法ほうほう来らい解かい决零样本IE任にん务。我わが们提出ていしゅつ一いち个问题：不ふ经过训练来らい实现零れい样本信しん息いき抽取是ぜ否ひ可か行ぎょう？我わが们将零れい样本IE任にん务转变为一个两阶段框架的多轮问答问题（Chat IE）,并在三さん个IE任にん务中广泛评估了りょう该框架か：实体关系三さん元げん组抽取、命名めいめい实体识别和わ事件じけん抽取。在ざい两个语言的てき6个数据すえ集しゅう上じょう的てき实验结果表明ひょうめい，Chat IE取得しゅとく了りょう非常ひじょう好このみ的てき效果こうか，甚至在ざい几个数すう据すえ集しゅう上じょう（例れい如NYT11-HRL）上じょう超ちょう过了全ぜん监督模型もけい的てき表ひょう现。我わが们的工作こうさく能のう够为有限ゆうげん资源下かIE系けい统的建立こんりゅう奠定基さだもと础。

UPDATE： we use the official api, the tool becomes more faster!!! if the key exceed limits please tell us.

NOTICE： The response speed depends on the official openai chatgpt api. (sometimes, the official is too crowded and the speed will be slow or the chatgpt will be overloaded.) Moreover, you better use your own openai key because if our default account is used by multiple people at the same time, the account may be overloaded.

NOTICE: because official api is not available in domestic, so we use api from revChatGPT and v1 version. But it's too slow, so we advise you use the tool offline for study. We will update the api further in the future (TODO).

we also provide a IE tool based on GPT3.5, you can see in GPT4IE

ChatIE (Zero-Shot Information Extraction via Chatting with ChatGPT) is a open-source and powerful IE tool demo. Enhanced by ChatGPT and prompting, it aims to automatically extract structured information from a raw sentence and make a valuable in-depth analysis of the input sentence. Harnessing valuable structured information helps corporations make incisive and business–improving decisions.

We support the following functions:

This task aims to extract triples from plain texts, such as (China, capital, Beijing) , (《如懿传》, 主演しゅえん, 周しゅう迅).

• sentence: a plain text.
• relation type list (rtl)* : {'relation type 1': ['subject1', 'object1'], 'relation type 2': ['subject2', 'object2'], ...}

PS: * denote optional, we set default value for them. But for better extraction, you should specify the three list according to application scenarios.

sentence: Four other Google executives the chief financial officer , George Reyes ; the senior vice president for business operations , Shona Brown ; the chief legal officer , David Drummond ; and the senior vice president for product management , Jonathan Rosenberg earned salaries of $ 250,000 each .
rtl: default, see file "default-types"
ouptut:

sentence: 第だい五ご部ぶ：《如懿传》《如懿传》是ぜ一部古装宫廷情感电视剧，由よし汪ひろし俊しゅん执导，周しゅう迅、霍建华、张钧甯、董ただし洁、辛からし芷蕾、童わらわ瑶よう、李り纯、邬君梅うめ等とう主演しゅえん。
rtl: default, see file "default-types"
ouptut:

This task aims to extract entities from plain texts, such as (LOC, Beijing) , (人物じんぶつ, 周しゅう恩来おんらい).

• sentence: a plain text.
• entity type list (etl)* : ['entity type 1', 'entity type 2', ...]

sentence: James worked for Google in Beijing, the capital of China. etl: ['LOC', 'MISC', 'ORG', 'PER']
ouptut:

sentence: 中国ちゅうごく共ども产党创立于中华民国大こくだい陆时期き，由ゆかり陈独秀和しゅうわ李り大だい钊领导组织。
etl: ['组织机つくえ构', '地点ちてん', '人物じんぶつ']
ouptut:

This task aims to extract event from plain texts, such as {Life-Divorce: {Person: Bob, Time: today, Place: America}} , {竞赛行ぎょう为-晋すすむ级: {时间: 无, 晋すすむ级方: 西北せいほく狼おおかみ, 晋すすむ级赛事ごと: 中ちゅう甲かぶと榜首之の争そう}}.

• sentence: a plain text.
• event type list (etl)* : {'event type 1': ['argument role 1', 'argument role 2', ...], ...}

sentence: Yesterday Bob and his wife got divorced in Guangzhou.
etl: default, see file "default-types"
ouptut:

sentence: 在ざい2022年ねん卡塔尔世界かい杯はい决赛中ちゅう，阿おもね根ね廷以点てん球だま大だい战险胜法国こく。
etl: default, see file "default-types"
ouptut:

react+flask

1. cd front-end and Run npm install to download required dependencies.
2. Run npm run start. ChatIE should open up in a new browser tab.
3. cd back-end and Run python run.py.
4. note: node-version v14.17.4 npm-version 9.6.0
5. you may need to configure proxy on your machine.

We are committed to improving our project and providing you with the best possible experience. To achieve this, we will collect your data to help us understand how you interact with our project and identify areas for improvement. We value the privacy and security of your data and ensure the data only for the purposes of improving our project.

Checkout this paper arxiv: 2302.10205

@article{wei2023zero,
  title={Zero-Shot Information Extraction via Chatting with ChatGPT},
  author={Wei, Xiang and Cui, Xingyu and Cheng, Ning and Wang, Xiaobin and Zhang, Xin and Huang, Shen and Xie, Pengjun and Xu, Jinan and Chen, Yufeng and Zhang, Meishan and others},
  journal={arXiv preprint arXiv:2302.10205},
  year={2023}
}