.png){kind=icon}
.png){kind=icon}
Return
NDMKG: a multimodal knowledge graph for nuclear data
.png){kind=icon}
PRE
AI
Y
R
F
L
X
X
DOI：10.1007/s41365-026-02029-7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Nuclear data are a fundamental resource underpinning nuclear science research and technological applications. Currently, nuclear data are distributed across multiple databases and encoded in heterogeneous formats, which complicates data integration and limits efficient knowledge mining and downstream applications. The construction of a nuclear data knowledge graph can effectively integrate heterogeneous data from multiple sources and provide support for applications such as question answering, retrieval, and recommendation. However, existing knowledge graph studies in nuclear science and technology remain largely limited to textual information, which can result in incomplete knowledge representation and missing semantic associations. To this end, this paper proposes a nuclear data multimodal knowledge graph (NDMKG) construction method. First, the ontology structure of NDMKG is designed and multimodal data is obtained from authoritative sources. For text data, a rule-based approach is used to extract structured knowledge; for image data, object detection, relation prediction, and optical character recognition techniques are integrated to achieve automatic extraction of entities, relationships, and attributes. Subsequently, the fusion of image and text knowledge is achieved by entity alignment, and the results are stored in a graph database to complete the construction of NDMKG. To enable fine-grained entity and relation extraction from decay scheme image data, this paper proposes a rotated object detection model, Rotated R-CNN, and an image relation prediction model, ImgRel. The experimental results show that Rotated R-CNN outperforms existing models on the constructed decay-scheme dataset, while ImgRel achieves optimal performance through multiple feature fusion. These results validate the effectiveness of the proposed models in knowledge extraction for decay schemes. Finally, a visualization system is developed based on NDMKG to support knowledge querying and filtering. This system provides an application interface for multimodal exploration of nuclear data.
Keywords:
Nuclear data
Multimodal knowledge graph
Image understanding
Knowledge extraction
Journal
IF:
3.8
Papers:
2.1K
Citations:
3.4K
Organization
Cited Papers