Computational Analysis of Thai Plant Diseases: A Preliminary Investigation Based on a Q&A Dataset

¹ Faculty of Science, Maejo University, No. 63 Moo 4, Nong Han Subdistrict, San Sai District, Chiang Mai, Thailand, 50290
² Faculty of Applied Science, King Mongkut’s University of Technology North Bangkok, 1518 Pracharat 1 Rd., Wongsawang, Bangsue, Bangkok, Thailand, 10800

^* Corresponding author: porawat.v@sci.kmutnb.ac.th

Abstract

Plant diseases pose a significant threat to Thailand's agricultural productivity and food security. This study leverages a domain-specific Thai-language Q&A dataset of 3.000 entries to analyze prevalent plant disease concerns and develop an Al-powered diagnostic assistant. Through fine-tuning a GPT-2 Thai language model (flax-conmiunity/gpt2-base-thai). we achieve robust performance with a training loss convergence to 0.20 (97.5% improvement) and evaluation metrics demonstrating high semantic accuracy (0.75 similarity) and linguistic coherence (11.50 perplexity) on a 500-sample test set. Natural Language Processing techniques—including topic modeling and keyword extraction—reveal key insights: (1) fungal diseases in rice dominate farmer inquiries, (2) early symptom identification for durian diseases is frequently misunderstood, and (3) prevention strategies for cassava mosaic virus are under-discussed. The fine-tuned model shows strong alignment with expeit knowledge (BLEU score: 0.50) while preserving meaning across paraphrased responses, highlighting its potential for scalable agricultural extension services. This work provides a blueprint for deploying NLP solutions in low-resource languages, emphasizing the value of domain-specific fine-tuning to bridge the gap between technical knowledge and farmer accessibility. Our results advocate for integrating such models into mobile advisory platforms to combat misinformation and strengthen Thailand's plant disease resilience.