Discord Bot Design for Hate Speech Sensor Using Convolutional Neural Networks (CNN) Method

Discord is growing in popularity, makes hard for an admin of Discord server maintaining their member in their everyday chat activity in their server. This no longer an issue if there is Discord bot that can detect hate speech feature in text message that member send and automatically censor them. The classifier for this experiment is using Convolutional Neural Network (CNN) method. The dataset for training and validation model are containing total 6 category of hates speech, abusive language, religion, race, gender, physical, and non-hate speech. The Discord bot program only can classify a message in Indonesian language. The dataset used for training and validation models was obtained from Kaggle and for additional data taken from Discord server messages totaling 18,986 sentences which will be divided by 80% training data and 20% test data. The final results of the training model experiment, this CNN model can classify test data with an average precision value of 89%, 90% recall, and 88,33% F1 score. The CNN model is integrated into a bot application which will be tested on messages sent from the test Discord server. Out of 279 messages, the designed Discord bot can obtain an accuracy of 70.6%.