What is ChatCEA?
It is a knowledge-driven intelligent service agent for controlled environment agriculture that uses LLMs with retrieval and domain-specific knowledge.
Why this work?
CEA growers work with distributed manuals, scientific papers, quantitative parameters, and evolving production decisions. General search and general-purpose language models often lack the context needed for precise, actionable answers.
Approach
ChatCEA combines a two-source knowledge base built from management manuals and journal papers, a retrieval module for contextual semantic search, and a knowledge-integration module for unstructured and quantitative information.
The work also establishes a CEA-specific benchmark spanning major production subdomains.
Key evidence
1,025Question-answer pairs in the CEA-specific benchmark
0.71Answer correctness, versus 0.41 for GPT-4 and 0.44 for Google Search in the reported evaluation
4.3Reported LLM-based evaluation score, versus 3.8 for GPT-4 and 2.5 for Google Search
Questions this work answers
What does its benchmark measure?
It supports systematic evaluation of qualitative and quantitative question answering across major CEA production subdomains.