Multilabel Sentiment Analysis Of Hate Speech Using The Combination Of Indo Bert Lite And Bidirectional Lstm-Cnn Methods With Grid Search Hyperparameter Optimization

According to a survey conducted by the Human Resources Research and Development Agency Kominfo in 2023, there were around 60,4% of social media users in Indonesia. However, there were negative messages or opinions on social media that could be hate speech. To classify the sentiment of netizens, whether it contains hate speech content, it can be seen from the terms posted. The labels used are multilabel based on emotion wordsThis study aims to conduct a multilabel sentiment analysis on hate speech. This study used a combination of pretrained indoBert models, BiLSTM-CNN and Hyperparameter Grid Search. The dataset used to train the model is a collection of 13,169 Indonesian-language tweets with 12 labels. The test results with the combination of the models already mentioned, obtained the highest hamming score in the test set was 72.61% with the best hyperparameter being a batch size of 90, with an epoch of 50, a dropout rate model of 0.2, a neuron model of 30, and the Adamax optimizer.