Automated Domain Question Mapping (DQM) with Educational Learning Materials

TL;DR

This work tackles automatic construction of pedagogy-oriented knowledge representations by introducing Domain Question Maps (DQMs), which structure domain knowledge as questions rather than concepts. It leverages QA datasets (KhanQ, SQuAD) to fine-tune question-generation models, and a specificity classifier to establish hierarchical relationships among generated questions; a graph post-processing pipeline using semantic similarity and edge weights $w_{ij} = \lambda \eta_{ij} + (1 - \lambda) \xi_{ij}, \lambda \in [0,1]$, plus a Maximum Spanning Tree (MaxST) via Kruskal, yields navigable DQMs. The results show encoder-decoder PLMs generally outperform decoder-only models for QG, with datasets and models varying in performance, while the relationship classifier achieves robust macro F1 (~0.90) on held-out data. A qualitative analysis on a real textbook demonstrates coherent question sequences along learning paths, indicating potential for personalized and adaptive learning applications. Overall, the approach offers a data-efficient, scalable pathway to generate structured, instructional question maps that support targeted learning objectives and domain understanding.

Abstract

Concept maps have been widely utilized in education to depict knowledge structures and the interconnections between disciplinary concepts. Nonetheless, devising a computational method for automatically constructing a concept map from unstructured educational materials presents challenges due to the complexity and variability of educational content. We focus primarily on two challenges: (1) the lack of disciplinary concepts that are specifically designed for multi-level pedagogical purposes from low-order to high-order thinking, and (2) the limited availability of labeled data concerning disciplinary concepts and their interrelationships. To tackle these challenges, this research introduces an innovative approach for constructing Domain Question Maps (DQMs), rather than traditional concept maps. By formulating specific questions aligned with learning objectives, DQMs enhance knowledge representation and improve readiness for learner engagement. The findings indicate that the proposed method can effectively generate educational questions and discern hierarchical relationships among them, leading to structured question maps that facilitate personalized and adaptive learning in downstream applications.

Figures (2)

• Figure 1: Results of graph post-processing. Figure (a) shows the t-SNE projection of question embeddings using Sentence-BERT, figure (b) shows the result of edge selection with a weight threshold ($\tau=0.7$), and figure (c) shows the final graph after pruning the graph using the MST algorithm.
• Figure 2: Sample paths over the final question graph after applying the MST algorithm. The question samples show the specificity relationship between the questions along the path.