Offensive Language Detection using Artificial Neural Network

摘要

Governments and social media providers put an effort to tackle offensive, abusive, and profanity in social media as an abuse of speech freedom. Considering the number of Internet user in Indonesia and the conflict caused by offensive content about religion, race, and inter-group issues in Indonesia, there is an urge to develop offensive content detection for posts written in Bahasa. This paper uses an artificial neural network model for not only classifying the words as (non)offensive words but also considering the structure of the sentence to get its context. The challenges are informal grammar and word abbreviation used in social media. Hence, there are noise elimination and normalization processes to address these challenges. The computer simulation results show excellence accuracy of 99.18% training, 94.28% validation, and 96.8% testing, only by utilizing the sigmoid activation function. This model can assist government enforcing the information and electronic transaction law and decreases the number of disputes due to aspiration freedom abuse in social media.