Temporal Knowledge Graph Reasoning With Historical Data Correlation in Agriculture

In terms of temporal dependencies of agricultural data, temporal knowledge graph reasoning is widely used to provide dynamic decision‐making for crop monitoring and disease prediction. Since periodicity, repeatability, and persistence always exist in large, various agricultural data, it is difficult to effectively capture the above important features for accurately predicting future events. To address the above limitations, we propose a novel recursive evolutionary network (NREN) to enhance temporal inference from structural and temporal aspects. A graph structural processing module is designed to capture potential relationships among consistent and accurate data, which is composed of self‐attention, a relational graph convolutional network, and a gated recurrent unit. In addition, historical correlation units and historical frequency units are designed to capture periodic patterns and repetitive features in the temporal dynamic processing module. Meanwhile, a gated recurrent unit is also integrated with the temporal dynamic processing module to represent temporal dynamics in parallel. Comparing with existing methods, experimental results demonstrate that NREN improves MRR and Hits@10 by an average of 7.43% and 4.77%, respectively, on five existing datasets. Especially, 1.86% and 3.06% represent an average increase in the Corn dataset.