Language dataset construction in Thai

Abstract

In today’s digital landscape, memes are a vital form of cultural expression, especially in Thailand where over two-thirds of the population is active on social media. Thai memes, which intricately blend tonal language, scripts, and cultural nuances, offer both humor and social commentary. These complexities, however, challenge automated analysis due to their nuanced nature. This study addresses these challenges by developing a neural network tailored to preserve the cultural and contextual integrity of Thai memes. The model combines Convolutional Neural Networks (using ResNet-18) for image recognition and Optical Character Recognition (OCR) for text extraction, trained on a dataset of manually labeled Thai memes collected through Selenium. This approach enables the network to accurately categorize memes and identify harmful content by capturing nuanced cultural references and detecting hate speech. The findings demonstrate the model’s effectiveness in making accurate predictions, presenting a significant advancement in culturally aware AI systems for content moderation. This study not only offers insights into the subtleties of Thai digital culture but also sets the groundwork for future research to enhance the robustness and cultural sensitivity of meme analysis algorithms.