Enhancing Toxic Comment Detection with BiLSTM-Based Deep Learning Model

Enhancing the ability to identify abusive comments is essential, as it addresses the growing issue of online toxicity while fostering healthier interactions. In this paper, a deep learning architecture is proposed using Bidirectional Long Short-Term Memory (BiLSTM) to enhance the identification of toxic comments with multiple labels. Due to its architectural characteristics, Bi-LSTM achieves high accuracy by thoroughly preprocessing data and effectively managing unknown words during FastText embeddings. It is especially effective at detecting subtle forms of intoxication, such as obscenity, insults, hate speech, and threats, by using contextual signals. Incorporating these techniques online may enable the automatic detection and deletion of toxic content, thereby promoting positive interactions. This model serves as a benchmark for improving the handling of classification problems with various labels through fine-tuning the neural network architecture.