MMKG-RDS: Reasoning Data Synthesis via Deep Mining of Multimodal Knowledge Graphs
#MMKG-RDS #reasoning data synthesis #multimodal knowledge graphs #fine‑grained knowledge extraction #path sampling #data quality scoring #Qwen3 #benchmark construction #arXiv Jan‑Feb 2026 #AI research
📌 Key Takeaways
- Introduces MMKG‑RDS, a framework for generating high‑quality reasoning data from multimodal knowledge graphs.
- Supports fine‑grained knowledge extraction and customizable path sampling.
- Employs multidimensional scoring to evaluate data quality.
- Validated on MMKG‑RDS‑Bench, covering five domains, 17 task types, and 14,950 samples.
- Fine‑tuning large language models (Qwen3) on a small set of synthesized data improves reasoning accuracy by 9.2 %.
- Generates distinct data that challenges models on tables and formulas, aiding complex benchmark creation.
📖 Full Retelling
This paper, authored by Lun Zhan, Feng Xiong, Huanyong Liu, Feng Zhang, and Yuhui Yin, proposes the MMKG‑RDS framework for reasoning data synthesis via deep mining of multimodal knowledge graphs. The work was submitted to arXiv on 27 February 2026 and presents a flexible method for extracting fine‑grained knowledge, customizing path sampling, and scoring data quality across multiple modalities. The authors aim to overcome current limits in long‑tail coverage, effectiveness verification, and interpretability of existing knowledge‑graph‑based data synthesis methods, thereby enhancing the reasoning abilities of domain models.
🏷️ Themes
Artificial Intelligence, Knowledge Graphs, Data Synthesis, Machine Learning Benchmarking, Multimodal Learning
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
--> Computer Science > Artificial Intelligence arXiv:2602.23632 [Submitted on 27 Feb 2026] Title: MMKG-RDS: Reasoning Data Synthesis via Deep Mining of Multimodal Knowledge Graphs Authors: Lun Zhan , Feng Xiong , Huanyong Liu , Feng Zhang , Yuhui Yin View a PDF of the paper titled MMKG-RDS: Reasoning Data Synthesis via Deep Mining of Multimodal Knowledge Graphs, by Lun Zhan and 4 other authors View PDF HTML Abstract: Synthesizing high-quality training data is crucial for enhancing domain models' reasoning abilities. Existing methods face limitations in long-tail knowledge coverage, effectiveness verification, and interpretability. Knowledge-graph-based approaches still fall short in functionality, granularity, customizability, and evaluation. To address these issues, we propose MMKG-RDS, a flexible framework for reasoning data synthesis that leverages multimodal knowledge graphs. It supports fine-grained knowledge extraction, customizable path sampling, and multidimensional data quality scoring. We validate MMKG-RDS with the MMKG-RDS-Bench dataset, covering five domains, 17 task types, and 14,950 samples. Experimental results show fine-tuning Qwen3 models (0.6B/8B/32B) on a small number of synthesized samples improves reasoning accuracy by 9.2%. The framework also generates distinct data, challenging existing models on tasks involving tables and formulas, useful for complex benchmark construction. The dataset and code are available at this https URL Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2602.23632 [cs.AI] (or arXiv:2602.23632v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2602.23632 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Feng Xiong [ view email ] [v1] Fri, 27 Feb 2026 03:10:51 UTC (1,084 KB) Full-text links: Access Paper: View a PDF of the paper titled MMKG-RDS: Reasoning Data Synthesis via Deep Mining of Multimodal Knowledge Graphs, by Lun Zhan and 4 other authors View PDF HTM...
Read full article at source