Knowledge Extraction
🔍 Knowledge Extraction: The Core Process of Deriving Knowledge from Data
Module Overview
Knowledge extraction is a critical step in building knowledge graphs, aiming to identify and extract semantically meaningful entities, relationships, and attributes from various data sources. The platform provides comprehensive knowledge extraction capabilities, encompassing four major functional modules: Concept Definition, Relationship Configuration, Structured Extraction, and Unstructured Extraction. These tools help users transform raw data into structured knowledge suitable for reasoning, analysis, and downstream applications.
This chapter details the operational workflows and usage methods for each functional module.
1. Concept Configuration
🧠 Functionality
The Concept Configuration module defines the fundamental semantic units—concepts—within the knowledge system, such as "Person," "Location," or "Organization." These concepts provide the foundational semantic framework for subsequent knowledge extraction and graph construction.
Workflow:
- Navigate to Knowledge Extraction → Concept Configuration;
- Click "Add" to enter the concept name and configure basic information such as color (used for visual representation of entities in the knowledge graph);
Existing concepts can be modified or deleted using the "Edit" or "Delete" buttons;
Supports keyword-based search for defined concepts;
All operations take effect in real time and are traceable via logs.
2. Relationship Configuration
⚙️ Functionality
The Relationship Configuration module defines association rules between different concepts, such as "Birthplace" or "Employed At," ensuring the system can correctly identify and establish semantic connections during the extraction process.
Workflow:
- Navigate to Knowledge Extraction → Relationship Configuration;
- Click "Add," select source and target concepts, and define the relationship type;
Relationship Configuration - Add
Existing relationships can be modified or deleted at any time;
Supports searching and filtering by relationship name or involved concepts.
3. Unstructured Extraction
📝 Functionality
The Unstructured Extraction module processes data sources without fixed formats, such as text files or web content. It leverages Natural Language Processing (NLP) techniques to automatically recognize and extract entities, events, and relationships.
Workflow:
- Navigate to Knowledge Extraction → Unstructured Extraction;
- Click "New Task," enter a task name, import knowledge files from the Knowledge Center, and import relevant triples for this extraction task, then save the task;
- Click "Execute" to start the task. The system will automatically run the extraction process—please wait for completion;
- Upon completion, click "Extraction Results" to view the output;
- Click on graph nodes to view detailed information and perform manual review. After approval, click "Publish" to release the graph into the final, unified knowledge graph, enabling downstream knowledge applications.
4. Structured Extraction
🗃️ Functionality
The Structured Extraction module extracts knowledge from structured databases. By mapping concepts and relationships, it generates standardized triple data.
Workflow:
- Navigate to Knowledge Extraction → Structured Extraction;
- Create a new task, enter a task name, select a data source, and import the data tables to be extracted;
Before performing structured extraction, ensure that your data source is configured. Refer to Data Source Configuration for instructions.
- Map table fields to corresponding concepts, attributes, and relationships.
(Tip: Data table → Concept; Table column → Attribute; Join table → Relationship)
The extraction process, manual review, and publishing workflow for structured data are the same as for unstructured data.
Note: Both structured and unstructured extraction tasks require a super administrator account to enable the system's scheduled tasks: Structured Task Extraction and Unstructured Task Extraction. Without these enabled, extraction tasks may remain in an "executing" state indefinitely.
✅ Summary
The Knowledge Extraction module forms the essential foundation for building knowledge graphs, covering the complete workflow—from concept definition and relationship configuration to extraction from structured and unstructured data. Whether processing textual documents or database records, the platform delivers efficient and accurate knowledge extraction capabilities, empowering enterprises to achieve intelligent knowledge management.