Offensive language and hate speech detection using deep learning in football news live streaming chat on YouTube in Thailand

Today, hate speech is frequently seen on Thai social media platforms such as Facebook, Twitter, and even online video platforms such as YouTube. In live video broadcasts of football news, for example, some Thais expressed hate speech toward opposing football fans and players. This paper presented offensive language and hate speech detection for Thai in YouTube live streaming chat with transformer-based language models by using five BERT models, including BERT, XLM-RoBERTa, DistilBERT, WangchanBERTa, and TwHIN-BERT, which were trained with multilingual languages as well as Thai. In the data labeling process, a two-step data labeling procedure was developed. The first stage involved automated data labeling utilizing the WangchanBERTa model, and the second stage involved manual data labeling conducted by the researchers. We developed text classification models using 11 different positive and negative class ratio datasets to get the most efficient model. In terms of recall and F1 score, the results showed that XLM-RoBERTa performed the best. It yielded an average recall and F1 score of 0.9669 and 0.9530, respectively. However, neither of the five models has significantly different performance. When considering the purpose of the application, DistilBERT is most appropriate. Because of its similar performance to XLM-RoBERTa, it has smaller model sizes and works faster.