Uie Base

🚀 UIE Information Extraction Model (Pytorch)

The UIE Information Extraction Model provides a unified framework for various information extraction tasks, enabling efficient and flexible data processing.

🚀 Quick Start

The UIE (Universal Information Extraction) model is a powerful tool for information extraction tasks. It offers a unified framework that can handle multiple types of extraction tasks, including entity extraction, relation extraction, event extraction, and sentiment analysis.

✨ Features

• Unified Framework: UIE (Universal Information Extraction) is a unified framework for information extraction proposed by Yaojie Lu et al. in ACL-2022. It unifies the modeling of tasks such as entity extraction, relation extraction, event extraction, and sentiment analysis, and enables good transfer and generalization capabilities between different tasks.
• Powered by PaddleNLP: To facilitate the use of UIE's capabilities, PaddleNLP trained and open-sourced the first Chinese general information extraction model UIE based on the ERNIE 3.0 knowledge-enhanced pre-trained model.
• Versatile and Adaptable: The model supports key information extraction without limitations on industry domains and extraction targets, enabling zero-shot rapid cold start and excellent few-shot fine-tuning capabilities to quickly adapt to specific extraction targets.

💻 Usage Examples

Basic Usage

from transformers import AutoModel, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("xusenlin/uie-base", trust_remote_code=True)
model = AutoModel.from_pretrained("xusenlin/uie-base", trust_remote_code=True)

schema = ["时间", "选手", "赛事名称"]  # Define the schema for entity extraction
print(model.predict(tokenizer, "2月8日上午北京冬奥会自由式滑雪女子大跳台决赛中中国选手谷爱凌以188.25分获得金牌！", schema=schema))

schema = {'竞赛名称': ['主办方', '承办方', '已举办次数']}  # Define the schema for relation extraction
model.set_schema(schema)
print(model.predict(tokenizer, "2022语言与智能技术竞赛由中国中文信息学会和中国计算机学会联合主办，百度公司、中国中文信息学会评测工作委员会和中国计算机学会自然语言处理专委会承办，已连续举办4届，成为全球最热门的中文NLP赛事之一。"))

schema = {'地震触发词': ['地震强度', '时间', '震中位置', '震源深度']}  # Define the schema for event extraction
model.set_schema(schema)
print(model.predict(tokenizer, "中国地震台网正式测定：5月16日06时08分在云南临沧市凤庆县(北纬24.34度，东经99.98度)发生3.5级地震，震源深度10千米。"))

schema = {'评价维度': ['观点词', '情感倾向[正向，负向]']}  # Define the schema for opinion extraction
model.set_schema(schema)
print(model.predict(tokenizer, "店面干净，很清静，服务员服务热情，性价比很高，发现收银台有排队"))

schema = "情感倾向[正向，负向]"  # Define the schema for opinion extraction
model.set_schema(schema)
print(model.predict(tokenizer, "这个产品用起来真的很流畅，我非常喜欢"))

schema = ['法院', {'原告': '委托代理人'}, {'被告': '委托代理人'}]  # Define the schema for opinion extraction
model.set_schema(schema)
print(model.predict(tokenizer, "北京市海淀区人民法院\n民事判决书\n(199x)建初字第xxx号\n原告：张三。\n委托代理人李四，北京市 A律师事务所律师。\n被告：B公司，法定代表人王五，开发公司总经理。\n委托代理人赵六，北京市 C律师事务所律师。"))

📚 Documentation

For more detailed information, please refer to the PaddleNLP repository.

📄 License

This project is licensed under the Apache-2.0 license.