Integration of AI with Knowledge Graphs for Smart Agriculture applications

This thesis explores the integration of large language models (LLMs) with semantic web technologies to enhance weather-aware decision-making in smart agriculture. The study focuses on bridging natural language input with structured environmental data by converting user queries into SPARQL using a GPT-based LLM. Historical weather data, including daily minimum, maximum, and average air temperatures, was collected from two publicly available weather APIs, demonstrating the role of semantic technologies in achieving data interoperability. These datasets were semantically modeled using the SOSA ontology and stored in a SPARQL Event Processing Architecture (SEPA) knowledge graph, enabling real-time querying over RDF triples. To evaluate the effectiveness of this approach, a prototype system was developed to process user questions in natural language and return weather observations by executing the corresponding SPARQL queries. The system was tested using a one-year dataset from both APIs for the De Bilt region (Netherlands), allowing for validation of query results and analysis of vocabulary alignment and model consistency. The findings demonstrate the potential of combining LLMs with semantic graph infrastructures to improve accessibility, usability, and interpretability of environmental data in agricultural applications.