Abstract:

Aiming at the problems of data enrichment, difficult knowledge transformation, high data barriers of existing tools, missing semantics, and insufficient flexibility in the era of agricultural big data, this study designs and develops an online analysis engine for agricultural scientific data. The engine adopts a layered architecture, including a user interaction layer, an intelligent workflow engine, a knowledge base and state management module, and a containerized execution layer. The core innovations of the engine: constructing a metadata-driven mechanism with data state descriptors and operator capability images, proposing a hybrid recommendation model of SC-MPARank, and designing a dynamic and evolvable pipeline with a domain semantics orientation. The engine achieves ‘continuous learning - real-time inference’ through knowledge graph, combining the flexibility of a general platform, the expertise of an expert system and the automation capability of AutoML, and can intelligently organize the execution of existing algorithms. The engine has been practically applied to three major scenarios of breeding, cultivation, and agricultural green development, effectively reducing the technical threshold, improving the efficiency and reliability of data-to-decision transformation, and providing a practical agricultural scientific data analysis tool for smart agriculture.

Key words: agricultural big data, on-line analysis engine, SC-MPARank hybrid recommendation model, knowledge graph