HKderived

HKderived is a framework for knowledge derivation and representation designed to produce hierarchically organized knowledge structures by deriving new concepts and relations from a base dataset. It prioritizes interpretable derivations that can be traced step by step, combining symbolic rules with statistical signals to support both explainability and scalability. The resulting structure, often represented as a directed derivation graph, consists of nodes representing concepts and edges denoting derivational steps or dependencies between them.

The term originated in the field of knowledge representation during collaborative work at the Horizon Knowledge Lab (HKL) in the late 2010s. The 'HK' in HKderived signals hierarchical knowledge, while 'derived' highlights the procedural generation of new elements from existing data. The framework has since been described in several position papers and method papers outlining its aims, core components, and evaluation methods.

HKderived combines rule-based transformation with probabilistic weighting. Each derivation step applies a finite set of rules to a current subgraph to produce an expanded subgraph; probabilities reflect evidence from data, context, and prior steps. The framework enforces consistency constraints to prevent cycles when used in acyclic configurations and supports multiple interrelated derivation strategies that can be composed in sequence.

Applications span knowledge graph construction, ontology evolution, information extraction, and educational content organization. For example, starting from a corpus of animal facts, HKderived can derive higher-level categories such as 'mammal' and 'vertebrate,' linking them to properties like 'has fur' and 'gives birth to live young' through rule-guided inferences and statistical validation.

See also Knowledge graph, Ontology engineering, Derivation system, Knowledge representation. References to HKderived appear in technical reports and conference papers from HKL-affiliated researchers, and in later comparative studies evaluating derivation-based representations.