Abusive Comment Detection in Social Media with Bidirectional LSTM Model
S. Akila Agnes, A. Arun Solomon, D. Joseph Charles Tamilmaran · 2023
Social media is becoming more vulnerable to problems like personal attacks and undesirable behaviour like cyberbullying. Online abusive language includes abusive comment, indiscriminate slang, harsh language, and vulgarity, which serve as an instrument for very intense and cruel cyber abuse. The abusive comments on other people’s thoughts and beliefs can be hurtful and demeaning to those who express their views on social media. Manually reviewing each comment takes a significant amount of time and effort to determine which comments to delete. Therefore, an automatic process of identifying and preventing negative posts would not only save time but also give users comfort on social media platforms. In this work, a bidirectional long short term memory (BIS TM) model is implemented to identify abusive Twitter comments by classifying them into offensive and non- offensive comments. And the effectiveness of machine learning based abusive comment detection models with TF-IDF and Word2vec text representation techniques are compared with the proposed BLS TM model. The performance of abusive comment recognition models are evaluated using 10-fold cross validation with various performance metrics such as AUC, precision, recall, fl score, and accuracy. The experimental results showed that the best performance was yielded by the deep learning model BLS TM on the offensive language identification dataset(OLID).