In a groundbreaking development poised to revolutionize the agriculture sector, researchers have harnessed the power of large language models (LLMs) to construct domain-specific knowledge graphs at a fraction of the traditional cost. Published in *Nongye tushu qingbao xuebao*, the study led by SHI Zhongyan and colleagues from the Agricultural Information Institute, Chinese Academy of Agricultural Sciences, introduces a novel method for building knowledge graphs using DeepSeek, an open-source LLM. This innovation addresses longstanding challenges in knowledge engineering, such as high dependency on expert rules and the prohibitive costs of manual annotation.
The research team proposed a semantic understanding-enhanced, cue-engineered domain knowledge extraction technology system. This system leverages DeepSeek’s advanced textual reasoning capabilities to intelligently extract and process multi-source heterogeneous data. “By combining DeepSeek with manual calibration, we ensure both the efficiency and accuracy of knowledge extraction,” explained LEI Jie, a co-author of the study. The methodology was tested on the entire pig industrial chain, identifying 21 core entities and their attribute relationships, with a focus on smart farming.
One of the most striking findings was DeepSeek-R1’s performance in zero-shot learning scenarios. The model achieved an impressive F1 value of 0.92 when recognizing the attributes of 161 diseases and 11 types of entities in pig disease control. This demonstrates the model’s potential for reusable knowledge extraction across various links in the industrial chain. “Our approach not only reduces costs but also ensures the scalability and adaptability of knowledge graphs,” added SUN Tan, another co-author.
The implications for the agriculture sector are profound. Knowledge graphs constructed using this method can integrate heterogeneous data from multiple sources, creating a unified industrial knowledge system. This system can be utilized for designing and validating intelligent application scenarios, ultimately promoting intelligent information processing in the pig industry and beyond. “This synergistic paradigm combines the strengths of deep learning and manual calibration, offering a practical reference for cost-effective knowledge graph construction,” noted ZHAO Ruixue, a key contributor to the research.
The study’s findings contribute significantly to the existing literature on knowledge engineering and offer a practical blueprint for leveraging LLMs in vertical domains. As the agriculture sector continues to embrace digital transformation, this research paves the way for more efficient, accurate, and cost-effective knowledge management solutions. The constructed knowledge map of the entire pig industrial chain is just the beginning, with potential applications extending to other agricultural domains.
This research, led by SHI Zhongyan and colleagues from the Agricultural Information Institute, Chinese Academy of Agricultural Sciences, and affiliated with the Key Laboratory of Agricultural Big Data and the Key Laboratory of Knowledge Mining and Knowledge Services in Agricultural Converging Publishing, marks a significant step forward in the field of agritech. As the industry continues to evolve, the integration of advanced technologies like DeepSeek will be crucial in driving innovation and efficiency.