Community-Based Innovation for Poverty Alleviation: A Text-Based Audit of Area-Based Research Funding in Thailand

Abstract

This study evaluates a Thai area-based research funding portfolio using an explainable multi-label text classification framework. The analysis used 629 project records containing project titles, innovation descriptions, funding-source metadata, year, and binary poverty-related labels. Seven labels with sufficient support were retained for supervised modeling: Lack_Skills, Soil_Issues, Water_Shortage, Tech_Adaptation_Issues, Disaster_Risk, Vulnerable_Groups, and Welfare_Access. Because the corpus was small, the project texts were short, and the label distribution was highly imbalanced, the study used sparse Term Frequency-Inverse Document Frequency (TF-IDF) representations and interpretable classical machine-learning models rather than deep learning. The best held-out model was Linear Support Vector Machine (SVM), achieving macro-F1 = 0.5391 and weighted-F1 = 0.7798 under stratified five-fold cross-validation. Logistic Regression was second-best with macro-F1 = 0.5114 and weighted-F1 = 0.7582. The fold-level difference between Linear SVM and Logistic Regression was statistically significant (paired t-test p = 0.0021, Cohen's d = 3.1539; McNemar p = 0.0027), although the small number of folds requires cautious interpretation of effect-size magnitude. Ablation analysis showed that unigram-only text slightly outperformed the full feature set, while title-only modeling substantially reduced performance. Explainable lexical features were dominated by terms related to poverty, soil, water, disaster, and data. The findings support a practical role for explainable text analytics in portfolio screening, funding governance review, and evidence-informed monitoring of poverty-oriented research portfolios.   Received: 19 April 2026 | Revised: 8 June 2026 | Accepted: 23 July 2026   Conflicts of Interest The authors declare that they have no conflicts of interest to this work.   Data Availability Statement The data that support the findings of this study are openly available on GitHub at https://github.com/pjarupunphol/Datasets/blob/main/Research_Funding_Data.zip.   Author Contribution Statement Wichidtra Sudjarid: Conceptualization, Methodology, Validation, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Project administration, Funding acquisition. Pita Jarupunphol: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.