@bert_tagalog
Multilabel Tagalog Hate Speech Classifier (MLTHSC) is a free to use AI-powered tool designed to classify hate speech in Tagalog text according to Age, Gender, Physical, Race, Religion, and Other categories. It uses a fine-tuned Tagalog Bidirectional Encoder Representation from Transformers (BERT) model to analyze text and a linear neural network to identify various categories of hate speech.
MLTHSC can detect hate speech across the following categories:
Target of hate speech pertains to one's age bracket or demographic
Target of hate speech pertains to gender identity, sex, or sexual orientation
Target of hate speech pertains to physical attributes or disabilities
Target of hate speech pertains to racial background, ethnicity, or nationality
Target of hate speech pertains to affiliation, belief, and faith to any of the existing religious or non-religious groups
Target of hate speech pertains other topic that is not relevant as Age, Gender, Physical, Race, or Religion
MLTHSC uses a machine learning model based on the BERT architecture fine-tuned on Tagalog hate speech data. When you input text and press the Classify button, the quantized model We used a quantized model for this demo because loading the original model would take time to load and perform classifications due to its larger size (500+ mb). The model being quantized means its size was reduced (100+ mb) so that it load and perform faster inference on phones while maintaining similar albeit reduced accuracy as the original. (hosted on Hugging Face ) will process the text and output the probability scores for each hate speech category. For this demo, we used a 30% threshold to determine if a category is present in the hate speech text.