Adapting Knowledge Graphs Representation for Storing Learning Problems in an Intelligent Programming Tutor: LOQI Format for Human-Readable Knowledge Graphs

In programming education, learning problems such as code tracing require intelligent tutoring systems (ITSs) to verify multi-step solutions and provide precise feedback. To do this, ITSs need formal representations that capture program entities, their relationships, and state transitions. A research gap arises for moderate-sized human-read (HR) execution-level knowledge graphs (KGs): unlike machine-processed or query-oriented graphs, they must be authored, inspected, and debugged as whole documents, yet mainstream textual serializations are poorly suited to that workflow. We present LOQI (Language for Ontology Query and Inference), an author-centric language and serialization format for compact KG representation. LOQI targets moderate-sized KGs that are used in automated reasoning and must be assessed by humans. Our evaluation of LOQI consists of two parts: technical and educational. Technically, we evaluate LOQI against RDF/Turtle, JSON-LD, and LPG/Cypher on readability-oriented structural metrics over a corpus of 20,973 CompPrehension learning-problem graphs, compare dedicated LOQI reasoning with generic backends, and report a preference survey. Educationally, we report a one-group pre-test–post-test classroom study involving 87 students who used a LOQI-based CompPrehension tutor for recursive code tracing. Wilcoxon signed-rank tests showed significant pre/post score changes; however, the single-group design does not support causal attribution of those changes to the tutor or to the LOQI format.