Related Experiment Video
Updated: May 12, 2026
07:35
A Knowledge Graph Approach to Elucidate the Role of Organellar Pathways in Disease via Biomedical Reports
Published on: October 13, 2023
A method for constructing a multi-modal knowledge graph of sheep disease based on a pre-trained model
Liu Jiahao1, Wang Fushun2,3, Yuan Wanzhe4,5
1College of Information Science and Technology, Hebei Agricultural University, Baoding, China. 20237060902@pgs.hebau.edu.cn.
Veterinary Research Communications
|May 11, 2026
Summary
This study introduces a novel method for constructing sheep disease multi-modal knowledge graphs, significantly improving diagnostic accuracy and model efficiency in animal husbandry. The approach enhances semantic representation for intelligent disease diagnosis with limited data.
Area of Science:
- Veterinary Science
- Artificial Intelligence
- Data Science
Background:
- The increasing complexity of animal husbandry knowledge highlights the need for advanced tools like multi-modal knowledge graphs (MMKG) for intelligent animal disease diagnosis.
- Current MMKG construction faces challenges including limited professional knowledge, scarce multi-modal data, and insufficient labeled samples, particularly in specialized fields like sheep disease diagnosis.
Purpose of the Study:
- To propose and validate a novel method for constructing a sheep disease multi-modal knowledge graph (MMKG).
- To address the challenges of data scarcity and improve the efficiency and accuracy of intelligent sheep disease diagnosis.
Main Methods:
- Construction of a text knowledge graph using RoBERTa+BiLSTM+CRF sequence labeling for multi-modal knowledge alignment.
- Cross-modal representation learning via a vision-language pre-training two-stream model with LoRA fine-tuning and Bayesian optimization for domain adaptation.
- Multi-modal knowledge alignment, storage, and visualization using the GraphXR tool and proposed rules.
Main Results:
- Achieved high image-text matching accuracy: 81.82% on a self-built sheep disease dataset and 93.11% on the EuroSAT dataset.
- Demonstrated significant improvements over pre-fine-tuning performance, with accuracy increases of 40.91% and 38.04%, respectively.
- Validated the effectiveness of fusing pre-trained models with fine-tuning for MMKG construction in small-sample, vertical domains.
Conclusions:
- The proposed method effectively constructs sheep disease multi-modal knowledge graphs, enhancing semantic representation and diagnostic capabilities.
- This approach significantly improves model efficiency and generalization ability, especially in data-scarce environments.
- Provides a robust knowledge base and technical support for intelligent sheep disease diagnosis applications.
