Thai Fake News Detection with LLM Integration for Web Application

This study presents a Thai fake news detection web application that employs an ensemble of deep learning models—Bi-LSTM, bert-base-thai, and mdeberta-v3-base —combined through a majority voting approach to enhance classification performance. Using this ensemble method, the system achieves up to 95\% accuracy in binary classification of news as Real or Fake. Rather than merely informing users of the classification result, the system also provides a summary of the article along with an explanation of the classification rationale. For these tasks, mT5-base is used for summarization and Typhoon-7B for generating reasoning. The system accepts input in the form of text, URLs, or images, retrieves related articles using SerpAPI and Crawl4AI, and delivers explanations and summaries in real time through a user-friendly web interface.