Dual index retrieval augmented generation for pedagogically aligned mathematics problem solving support in mathematics education
Discover Computing · 10.08.2026 · 29 (1)
DOI: 10.1007/s10791-026-10432-2 ↗Kurz gefasst
Fachartikel in der Zeitschrift Discover Computing, online erschienen am 10.08.2026. Verfasst von Haozhe Jiang und 2 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: The Education Programs of the National Social Science Fund of China.
Abstract (Originalsprache)
The use of large language models (LLMs) in mathematics education remains constrained by hallucinations and by weak alignment with classroom solution norms. Retrieval-augmented generation (RAG) can improve factual grounding, but standard retrieval pipelines often miss curriculum structure and may pass misleading near-neighbor examples to the generator. This study examines a dual-index RAG framework for high school mathematics that combines semantic vector retrieval in ChromaDB with a lightweight NetworkX knowledge graph, followed by similarity-based post-retrieval filtering and instructional prompting for stepwise explanations. On a curated corpus of Gaokao mathematics materials from 2000 to 2024, an evaluation on a 50-query retrieval benchmark shows that hybrid retrieval improves over the vector-only baseline, while the filtered full system is indistinguishable from the unfiltered hybrid condition under its original heuristic setting. The generation results indicate relative gains of 48.4% in faithfulness and 59.2% in pedagogical alignment over the vector-only baseline. A preliminary human–AI alignment check on 30 cases with two annotators produced an average correlation of $$r=0.84$$ and Cohen’s $$\kappa =0.76$$ . These findings should be interpreted as preliminary evidence from a small-scale benchmark, but they indicate that combining structural retrieval and pedagogically constrained prompting is a promising direction for mathematics problem solving support.
Förderung
- The Education Programs of the National Social Science Fund of China
Zitieren
APA
Jiang, H., Deng, Y., & Wang, Z. (2026). Dual index retrieval augmented generation for pedagogically aligned mathematics problem solving support in mathematics education. Discover Computing, 29(1). https://doi.org/10.1007/s10791-026-10432-2
Vancouver
Jiang H, Deng Y, Wang Z. Dual index retrieval augmented generation for pedagogically aligned mathematics problem solving support in mathematics education. Discover Computing. 2026;29(1). doi:10.1007/s10791-026-10432-2
BibTeX
@article{Jiang2026,
author = {Jiang, Haozhe and Deng, Yonghao and Wang, Zhaochun},
title = {Dual index retrieval augmented generation for pedagogically aligned mathematics problem solving support in mathematics education},
journal = {Discover Computing},
year = {2026},
volume = {29},
number = {1},
doi = {10.1007/s10791-026-10432-2}
}
EUVIMED