Instructor-Aligned Knowledge Graphs for Personalized Learning

Abdulrahman AlRabah; Priyanka Kargupta; Jiawei Han; Abdussalam Alawini

cs.AIFeb 19, 2026

Instructor-Aligned Knowledge Graphs for Personalized Learning

Authors: Abdulrahman AlRabah, Priyanka Kargupta, Jiawei Han, Abdussalam Alawini

Organizations: University of Illinois Urbana-Champaign, Siebel School of Computing and Data Science, Urbana, IL, USA

Abstract

Mastering educational concepts requires understanding both their prerequisites (e.g., recursion before merge sort) and sub-concepts (e.g., merge sort as part of sorting algorithms). Capturing these dependencies is critical for identifying students' knowledge gaps and enabling targeted intervention for personalized learning. This is especially challenging in large-scale courses, where instructors cannot feasibly diagnose individual misunderstanding or determine which concepts need reinforcement. While knowledge graphs offer a natural representation for capturing these conceptual relationships at scale, existing approaches are either surface-level (focusing on course-level concepts like "Algorithms" or logistical relationships such as course enrollment), or disregard the rich pedagogical signals embedded in instructional materials. We propose InstructKG, a framework for automatically constructing instructor-aligned knowledge graphs that capture a course's intended learning progression. Given a course's lecture materials (slides, notes, etc.), InstructKG extracts significant concepts as nodes and infers learning dependencies as directed edges (e.g., "part-of" or "depends-on" relationships). The framework synergizes the rich temporal and semantic signals unique to educational materials (e.g., "recursion" is taught before "mergesort"; "recursion" is mentioned in the definition of "merge sort") with the generalizability of large language models. Through experiments on real-world, diverse lecture materials across multiple courses and human-based evaluation, we demonstrate that InstructKG captures rich, instructor-aligned learning progressions.

Figures & tables

Figure 1. InstructKG extracts concepts from lecture chunks, classifies their pedagogical roles, and infers relationships. In this example, merge sort depends on recursion as a prerequisite, and sorting algorithms and divide-and-conquer are part of merge sort. Two lecture chunks about merge sort are converted into concept-role lists and then into a small knowledge graph with depends\_on and part\_of edges centered at merge sort. — Figure 1. InstructKG extracts concepts from lecture chunks, classifies their pedagogical roles, and infers relationships. In this example, merge sort depends on recursion as a prerequisite, and sorting algorithms and divide-and-conquer are part of merge sort. Two lecture chunks about merge sort are converted into concept-role lists and then into a small knowledge graph with depends_on and part_of edges centered at merge sort.

Figure 2. Overview of the InstructKG framework showing the three-phase pipeline. System architecture diagram showing document processing, evidence aggregation, and relation judgment phases

Course	Format	Lectures	Pages	Chunks
Database Systems	Slides	27	569	552
NLP	Slides + Notes	15	971	962
Algorithms	Lecture Notes	9	222	221

Table 1. Dataset statistics for the three evaluated courses.

Method	Model	Algorithms		NLP		SQL
Method	Model	Node	Triplet	Node	Triplet	Node	Triplet
EDC	Llama-3B	43.99 $\pm$ 43.62	19.53 $\pm$ 31.50	45.08 $\pm$ 45.63	13.03 $\pm$ 22.69	48.49 $\pm$ 46.88	11.25 $\pm$ 23.11
	Llama-8B	50.73 $\pm$ 44.62	13.56 $\pm$ 23.21	45.55 $\pm$ 46.22	14.43 $\pm$ 23.10	48.08 $\pm$ 47.19	13.16 $\pm$ 22.99
	Qwen-14B	53.44 $\pm$ 43.76	17.41 $\pm$ 26.00	52.70 $\pm$ 46.24	18.35 $\pm$ 24.68	53.15 $\pm$ 47.04	15.55 $\pm$ 23.71
KG-Gen	Llama-3B	43.61 $\pm$ 41.52	25.28 $\pm$ 36.13	49.65 $\pm$ 45.71	35.16 $\pm$ 38.33	52.56 $\pm$ 47.22	30.85 $\pm$ 39.03
	Llama-8B	52.99 $\pm$ 40.85	38.44 $\pm$ 38.41	53.38 $\pm$ 45.72	38.64 $\pm$ 36.70	53.35 $\pm$ 47.25	39.97 $\pm$ 38.13

Table 2. Node-level significance and triplet-level accuracy across datasets. Values are reported as mean

\pm

standard deviation (%). Bold indicates best; † indicates second best.

Q1	Q2	Q3
Question. Resources NOT accessed by users in Segment A. Average duration per resource.	Question. List open tickets, sorted by most recently updated first.	Question. Total hours per project (must use time logs as the source table).
Buggy solution: ⬇ SELECT ResourceId , AVG ( Duration ) AS AvgDur FROM AccessLog WHERE EXISTS ( SELECT 1 FROM UserProfile U WHERE U . UserId = AccessLog . UserId AND U . Segment = ’A’ ) GROUP BY ResourceId	Buggy solution: ⬇ SELECT TicketId , UpdatedAt FROM Tickets WHERE Status = ’Open’ ORDER BY UpdatedAt ASC	Buggy solution: ⬇ SELECT ProjectId , SUM ( Hours ) AS TotalHours FROM ProjectMembers GROUP BY ProjectId
Bug: Opposite logic ( EXISTS instead of NOT EXISTS )	Bug: Wrong sort direction (should be DESC )	Bug: Wrong source table (should use TimeLogs )
Error concept: WHERE	Error concept: ORDER BY	Error concept: FROM

Table 3. SQL Question–Concept Mapping. Red indicates error-related concepts. Dependency order: ORDER BY

\rightarrow

WHERE

\rightarrow

FROM .

Figure 3. Comparison of knowledge graph outputs: ours (left) vs. KGGen (right).

Figure 4. Comparison of knowledge graph outputs: ours (left) vs. KGGen (right).

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Component	Parameter	Value
Document Ingestion	Max tokens per chunk	8191
	Merge peers	True
Concept Extraction	Concepts per chunk	1–5 words
	Deduplication	Case-insensitive
Context Clustering	Embedding model	all-MiniLM-L6-v2
	UMAP $n_{\text{components}}$	15

Appendix

Table 4. Hyperparameters used across all pipeline components.

Personalized Learning

See all 18 papers

K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs
May 10, 2026Hao Liang, Qihan Lin, Zhaoyang Han +5AI in Education Knowledge Augmentation for Language Models
1Peking University · Institute for Advanced Algorithms Research, Shanghai · 3OriginHub Technology +1
ProPRL: Property-Aware Prerequisite Relation Learning in Educational Knowledge Graphs
Aug 4, 2026Xinghe Cheng, Jiapu Wang, Chaobo He +2Relation Extraction Graph Representation Learning
1Jinan University · 2Nanjing University of Science and Technology · 3South China Normal University +1
Detecting Knowledge Gaps from Conversational AI Interactions Using Curriculum Prerequisite Graphs
Jun 9, 2026Youssef Medhat, Junsoo Park, Ploy Thajchayapong +1AI in Education Educational Data Mining
Georgia Institute of Technology, Atlanta, GA, USA

May 10, 2026J/K move · Enter open · S save

K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs

Hao Liang, Qihan Lin, Zhaoyang Han, Xiaochen Ma+4

1Peking University · Institute for Advanced Algorithms Research, Shanghai · 3OriginHub Technology+1

AI in Education Knowledge Augmentation for Language Models Educational Assessment KG Construction Knowledge Representation Education+1

Open paper AskPDF arXiv

Large language models (LLMs) are increasingly used in K-12 education, yet existing benchmarks such as C-Eval, CMMLU, GaokaoBench, and EduEval mainly evaluate factual recall through exam-style question answering. Effective educational AI additionally requires curriculum cognition: understanding how knowledge is structured through prerequisite chains, concept taxonomies, experiment-concept links, and pedagogical sequencing. To address this gap, we introduce K12-KGraph, a curriculum-aligned knowledge graph extracted from official People's Education Press textbooks across mathematics, physics, chemistry, and biology from primary to high school. The graph contains seven node types (Concept, Skill, Experiment, Exercise, Section, Chapter, Book) and nine relation types covering taxonomy, prerequisite, association, verification, assessment, location, and order. Based on this graph, we construct two resources: (1) K12-Bench, a 23,640-question multi-select benchmark spanning five graph-derived task families (Ground, Prereq, Neighbor, Evidence, and Locate); and (2) K12-Train, a KG-guided supervised fine-tuning corpus of approximately 2,300 QA pairs synthesized from graph structure and node attributes. Experiments reveal substantial deficiencies in curriculum cognition: on K12-Bench, Gemini-3-Flash achieves only 57% exact match, while the best open-source model, Gemma-4-31B-IT, reaches 46%. Under a strictly matched 2,300-sample SFT budget on Qwen3-4B-Base and Llama-3.1-8B-Base, K12-Train consistently outperforms equally sized subsets from eight mainstream instruction-tuning corpora on both GaokaoBench and EduEval, demonstrating that curriculum-structured supervision is highly sample-efficient for educational tuning. We release the graph, benchmark, training data, and full construction pipeline.

Abstract

Personalized Learning

Explore similar work

K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs

ProPRL: Property-Aware Prerequisite Relation Learning in Educational Knowledge Graphs

Detecting Knowledge Gaps from Conversational AI Interactions Using Curriculum Prerequisite Graphs