zjunlp/knowlm-13b-ie
The zjunlp/knowlm-13b-ie model is a 13 billion parameter language model developed by zjunlp, specifically fine-tuned for information extraction (IE) tasks. It excels at named entity recognition (NER), relation extraction (RE), and event extraction (EE) by incorporating negative sampling during training to enhance its ability to output 'NAN' for non-existent relations. While optimized for precise information extraction, its general applicability is reduced compared to broader models. This model is designed for structured data extraction from text, supporting various IE templates and output formats.
Loading preview...
Overview
The zjunlp/knowlm-13b-ie is a 13 billion parameter model developed by zjunlp, specifically designed for information extraction (IE) tasks. It differentiates itself from the zjunlp/knowlm-13b-zhixi model by prioritizing information extraction utility, albeit with a trade-off in general applicability. This model was trained using a unique negative sampling approach, where approximately 10% of data from English and Chinese IE datasets were sampled, and then negative relations were randomly introduced. This method enhances the model's ability to output 'NAN' when a relation does not exist, improving its precision in IE.
Key Capabilities
- Specialized Information Extraction: Optimized for Named Entity Recognition (NER), Relation Extraction (RE), Event Extraction (EE), Event Type Extraction (EET), and Event Argument Extraction (EEA).
- Flexible Templating: Supports customizable instruction templates for various IE tasks, allowing users to define task descriptions, candidate label lists (
s_schema), and structured output formats (s_format). - Negative Sampling Integration: Training incorporates negative sampling to improve the model's robustness in identifying and explicitly marking non-existent entities or relations with 'NAN'.
- Structured Output: Provides tools and scripts to convert raw data into structured formats suitable for training and inference, including detailed examples for NER, RE, and EE tasks.
Use Cases
- Knowledge Graph Construction: Ideal for extracting entities, relations, and events to populate knowledge bases.
- Automated Data Annotation: Can be used to automatically annotate text with structured information for various downstream NLP tasks.
- Domain-Specific IE: Adaptable for specialized information extraction in fields like finance, law, or medicine by defining custom schemas.