The proposed approach, inspired by linguistic frame semantics, rule-based systems, and modern LLM architectures, attempts a human-in-the-loop methodology to empower individuals and organizations in curating reliable knowledge bases. The system is founded on a hierarchical concept model, which organises knowledge into Medium-level Concepts (MLCs), High-level Concepts (HLCs), Extractions, and userdefined Entities, interconnected by specific relationship types like HAS HLC, RELATED TO, HAS CHAIN, HAS ENTITY, and COMBINATION OF. This structure is implemented as a prototype using a Design Science Research (DSR) methodology, leveraging Python with FastAPI for the backend, SpaCy for natural language processing, Neo4j as the graph database, and a VueJS frontend.

Evaluation across use cases including Intelligent Entity Creation, Exploratory Navigation, and Comparison and Analysis of Texts demonstrated the prototype’s strengths. It outperformed LLMs and traditional tools by offering user-defined granularity, contextbased linking, and clear traceability, which are often absent in purely automated systems. Notably, for text analysis and comparison, the concept-based network identified more insightful phrases and deeper connections, providing a long-term, structured, and verifiable knowledge base. This system promotes human-machine cooperation, enabling users to build and maintain knowledge with verifiable connections and fostering innovation through explicit and discoverable relationships.