Intelligent Q&A for Transportation Engineering Integrating Course Knowledge Graphs and RAG: Identifying Weaknesses and Generating Personalized Learning Paths

Focusing on the complex knowledge structure, strong conceptual dependencies, and significant individual differences among learners in traffic engineering courses, this study proposes an intelligent question answering approach that integrates a course knowledge graph with retrieval augmented generation. The goal is to enhance professional consistency, evidence traceability, and learning support effectiveness in course level question answering. The method constructs a traffic engineering oriented knowledge graph through structured modeling of core course concepts, prerequisite relationships, and instructional resources. This knowledge graph is embedded into both the retrieval stage and the generation stage to impose semantic and structural constraints on the natural language question answering process. During question processing, student queries are first mapped to relevant nodes in the knowledge graph. Evidence texts are then retrieved from the course resource repository based on node relationships, forming a context set aligned with the course setting. On this basis, a knowledge constrained generation mechanism is applied to produce answers, ensuring consistency with course knowledge boundaries and instructional objectives. In addition, the interaction data between questions and knowledge mappings are further used to characterize learners' knowledge mastery states. This supports the identification of weak knowledge points and the generation of personalized learning paths. Comparative analysis demonstrates that the proposed method achieves more consistent advantages in retrieval relevance, evidence coverage, and knowledge node alignment. These results indicate that deep integration of course knowledge graphs with retrieval augmented generation improves the reliability and instructional value of intelligent question answering in traffic engineering education.

Paper

Full text

PDF

Intelligent Q&A for Transportation Engineering Integrating Course Knowledge Graphs and RAG: Identifying Weaknesses and Generating Personalized Learning Paths

Semantic Scholar · 2026

Abstract

Focusing on the complex knowledge structure, strong conceptual dependencies, and significant individual differences among learners in traffic engineering courses, this study proposes an intelligent question answering approach that integrates a course knowledge graph with retrieval augmented generation. The goal is to enhance professional consistency, evidence traceability, and learning support effectiveness in course level question answering. The method constructs a traffic engineering oriented knowledge graph through structured modeling of core course concepts, prerequisite relationships, and instructional resources. This knowledge graph is embedded into both the retrieval stage and the generation stage to impose semantic and structural constraints on the natural language question answering process. During question processing, student queries are first mapped to relevant nodes in the knowledge graph. Evidence texts are then retrieved from the course resource repository based on node relationships, forming a context set aligned with the course setting. On this basis, a knowledge constrained generation mechanism is applied to produce answers, ensuring consistency with course knowledge boundaries and instructional objectives. In addition, the interaction data between questions and knowledge mappings are further used to characterize learners' knowledge mastery states. This supports the identification of weak knowledge points and the generation of personalized learning paths. Comparative analysis demonstrates that the proposed method achieves more consistent advantages in retrieval relevance, evidence coverage, and knowledge node alignment. These results indicate that deep integration of course knowledge graphs with retrieval augmented generation improves the reliability and instructional value of intelligent question answering in traffic engineering education.

References (12)

11Predictive vertical CPU autoscaling in Kubernetes based on time-series forecasting with Holt-Winters exponential smoothing and long short-term memory2021
12U-niahACM Transactions on Information Systems

Similar papers

© 2026 NYSGPT2525 LLC