ENHANCING LOCAL SELF-GOVERNMENT INITIATIVES IN INDIAN DIGITAL AGRICULTURE THROUGH THE USE OF GRAPH CONVOLUTIONAL NETWORKS

Povzetek

The agricultural sector involves complex relationships among various entities such as crops, fertilizers, soil types, and seasons. Representing and managing these relationships in a meaningful and structured way is essential for efficient information retrieval and knowledge sharing. RDF-based Agricultural Knowledge Graphs (AKGs) have emerged as an effective solution for modelling these relationships; however, they are often static and lack support for advanced querying, semantic enrichment, and classification. This project presents a comprehensive system that enhances RDF-based AKGs using Graph Convolutional Networks (GCNs) for structural learning and optimization. The process begins with parsing raw RDF data in Turtle (.ttl) format and constructing directed graphs using NetworkX. These graphs are then converted into PyTorch Geometric objects for training a multi-layer GCN model, which generates meaningful node embeddings. A Top-K pooling technique is applied to filter redundant or insignificant nodes, resulting in a more concise and interpretable graph. The final, enriched graph is exported to Neo4j for semantic visualization and querying, enabling efficient extraction of agricultural insights. Experimental results demonstrate significant improvements in graph clarity, classification accuracy training: 98.22%, testing: 99.02%, and query effectiveness. This approach supports better data organization, simplifies knowledge retrieval, and strengthens the utility of AKGs in agriculture.