↗ arXiv ↗ Hugging Face ↗ Papers with Code
TL;DR#
Current knowledge extraction systems often struggle with diverse data formats, complex schemas, and error handling. Previous approaches often lack adaptability and require retraining for various tasks and datasets. This limits their effectiveness in real-world applications where data is messy and schema varies.
OneKE tackles these challenges with a novel multi-agent architecture. It uses a schema agent to preprocess data and generate appropriate schemas; an extraction agent to extract knowledge using LLMs, and a reflection agent to debug and correct errors using a case repository. This design allows OneKE to handle various data formats, adapt to different schemas (or lack thereof), and continuously learn from mistakes. The system’s dockerized design and open-source nature promote accessibility and reproducibility. The empirical evaluations demonstrate its effectiveness across different datasets and tasks.
Key Takeaways#
Why does it matter?#
This paper is important because it introduces OneKE, a novel and versatile system for knowledge extraction that addresses limitations of existing methods. Its dockerized design, schema-guided approach, and multi-agent architecture make it highly adaptable and efficient for various tasks and data types. The open-source nature of OneKE further enhances its accessibility and potential impact on the research community, paving the way for further improvements and applications in diverse fields.
Visual Insights#
🔼 OneKE is a dockerized schema-guided LLM agent-based knowledge extraction system. The figure illustrates its modular design, showcasing three core components: the Schema Agent (processes input data and schemas, handling various data types and formats like HTML and PDF), the Extraction Agent (extracts knowledge using LLMs, adapting to diverse tasks and LLMs), and the Reflection Agent (refines and corrects errors using a case repository). The system supports diverse domains (science, news, etc.) and utilizes a configurable knowledge base for schema management, debugging, and improvement. The diagram visually represents the flow of information through the system, highlighting the interaction between the agents and the knowledge base.
read the caption
Figure 1. The overview of the OneKE system, supporting various domains (science, news, etc.) and data (Web HTML, PDF, etc.).
Full paper#
