Semantic and Contextual Modeling for Malicious Comment Detection with BERT-BiLSTM