Natural Language Processing On Facebook Data For Optimized Disaster Response Strategies In Urdaneta City

Abstract

This study implements Natural Language Processing (NLP) on Facebook comment data to enhance disaster response in Urdaneta City, Philippines. Traditional disaster management often suffers from delayed, fragmented information. The developed system addresses this gap by applying an Extract-Transform-Load (ETL) framework integrated with an OpenAI GPT algorithm, enabling the real-time extraction and analysis of community sentiment and urgent help requests from Facebook comments. Key features include sentiment classification, emotion and intent detection, and named entity recognition, allowing rapid identification of disaster hotspots, resource needs, and distress signals. The model achieved 90.2% accuracy, a 94% F1 score, and average processing latency of 10.5 seconds, with robust adaptability across Filipino, Ilocano, and Taglish inputs (only 7% accuracy variance). Expert validation confirmed that 85% of flagged content directly aligned with actual disaster needs, while system error rates showed a bias toward over-alerting—a practical safety tradeoff in emergency settings. Dashboard tools revealed community sentiment trends and real-time operational backlogs, with most urgent requests emerging from comment threads. User acceptance, assessed via Technology Acceptance Model (TAM) surveys, was high: the majority found the system useful, easy to use, and beneficial for the community. Results highlight the value of leveraging AI-driven NLP on local social media to close feedback gaps, support early warning, and enable faster, data-driven disaster operations. This scalable, people-centered approach can serve as a template for other LGUs aiming to modernize their disaster response using accessible digital infrastructure.