Research progress on knowledge graph technology and its application in agriculture

An accurate and rapid analysis of massive data can be one of the most important steps for the comprehensive utilization of agricultural big data, particularly with the advent of the era of big data. Among them, the knowledge graph can be expected to represent complex domain knowledge using data mining, information processing, knowledge statistics, and graph drawing. The dynamic development of knowledge can also be revealed to provide a practical and valuable reference for complicated research. Therefore, the knowledge graph has attracted much attention in recent years, due mainly to the heterogeneous semantic network. Efficient management can be achieved in the things and their relationships in the real world. The efficient capability of information retrieval can be attributed to the storage structure of the knowledge graph, namely the directed graph. The knowledge graph can be applied to intuitively display the complex data under the background of the increasing data volume and complex structure in the agricultural field. Agricultural big data can be systematically analyzed for the high utilization of data value, and the mining of agricultural data rules, in order to promote the development of smart agriculture. The key technology of knowledge graph construction can dominate the knowledge graph research in the agricultural field. The agricultural knowledge graph should follow the general technology specification of knowledge graph construction. This review aims to explore the theoretical support of knowledge graphs in agricultural application. Firstly, the construction models of the knowledge graph were divided into three types: top-down, bottom-up, top-down and bottom-up combination. Among them, the top-down and bottom-up combination of construction model was the most commonly used with the more completed and flexible structure suitable for the knowledge graph construction in specific fields. Secondly, the key technologies of agricultural knowledge graph construction were summarized from five aspects: ontology construction, knowledge extraction, knowledge fusion, knowledge reasoning, knowledge graph storage and visualization. The progress of each aspect was then compared, including the technical difficulties, technological evolution, innovation and application exploration. It was found that Protégé tools and semi-automatic construction were widely adopted to construct the knowledge graph in the agricultural field. Furthermore, one of the most concerned research hotspots was knowledge extraction as the premise of construction. The best performance was obtained in the BERT-BiLSTM-CRF among the various knowledge extraction. The difficulty of knowledge graph construction was focused mainly on the terms recognition among agricultural subfields. Particularly, there was the great influence of the natural environment and climate on agricultural knowledge. The application research was also reviewed, including agricultural thematic literature metrology research, agricultural knowledge question and answer, agricultural information resources recommendation, and agricultural information retrieval, as the knowledge graph was gradually applied to the agricultural field. The ontology construction, knowledge extraction, knowledge graph storage, and visualization technologies were commonly used in the above application scenarios, but knowledge fusion and knowledge reasoning were rarely used, indicating the nonstandard knowledge graph construction under specific application backgrounds. Therefore, the agricultural knowledge graph should pay more attention to the cutting-edge construction technologies in the future, in order to innovate in combination with the characteristics of agricultural data. Finally, the future research trends of the knowledge graph can be expected to serve as the e-commerce recommendation for agricultural products. Two research directions were then proposed in the dynamic updating on the construction and correlation of knowledge graphs across the domain.