LLM-Based Tagging of Learning Resources with Evidence and Graph Constraints
An arXiv paper (2605.28483) introduces a comprehensive alignment pipeline that leverages a large language model (LLM) as a specialized tagger to connect educational resources with structured competency frameworks. This system dissects LMS materials—encompassing both instructional content and assessments—into pedagogical segments, identifies potential competencies from structured profiles enhanced by graph-based context, and employs the LLM to pinpoint the most pertinent competencies along with supporting evidence spans. The predictions are fine-tuned using the competency graph structure and compiled at the resource level. This method seeks to facilitate competency-driven searches and curriculum analytics, all while minimizing manual tagging efforts and enhancing transparency.
Key facts
- arXiv paper ID: 2605.28483
- Proposes an LLM-based tagging pipeline for learning resources
- Pipeline segments resources into pedagogical fragments
- Retrieves candidate competencies from structured profiles with graph context
- LLM selects competencies and provides evidence spans
- Predictions refined using competency graph structure
- Aggregated at resource level
- Aims to reduce manual tagging and improve transparency
Entities
Institutions
- arXiv