AUGMENTING ABUSIVE WORD IN SOCIAL MEDIA WITH WORD EMBEDDING

Abstract

The increase in the use of abusive language on social media lately is very bad. Many parties throw abusive words at each other against an object, either personal or group. Abusive words themselves can be in the form of sexism, attacking flaws or disabilities, and others. Activities on social media are now so negative that they do more harm than good. We use Word2vec and some algorithms to detect abusive words in hate speech on social media to see who’s the best algorithms so far that compatible work together with word2vec. First, we need to know the dataset we use from Kaggle.com. Then, for implementation, the dataset needs to be processed in data preprocessing, with steps such as word embedding, so that maximum results can be obtained. The final result of this project will be presented in a table of confusion matrix, and with this research, the calculated average F1 value is 86% and the accuracy rate is also 86%. So, with that result, we know that the final result is that the most suitable algorithm for this dataset is XGBoost, but the algorithm the most suitable with word2vec is KNearestNeighbor.