Proceedings of the 1st International Conference on Artificial Intelligence, Communication, IoT, Data Engineering and Security, IACIDS 2023, 23-25 November 2023, Lavasa, Pune, India

Research Article

Bidirectional LSTM with convolution for toxic comment classification

Download61 downloads
  • @INPROCEEDINGS{10.4108/eai.23-11-2023.2343140,
        author={Ashish Shinde and Pranav Shankar and Atul Atul and Srikari Rallabandi},
        title={Bidirectional LSTM with convolution for toxic comment classification},
        proceedings={Proceedings of the 1st International Conference on Artificial Intelligence, Communication, IoT, Data Engineering and Security, IACIDS 2023, 23-25 November 2023, Lavasa, Pune, India},
        publisher={EAI},
        proceedings_a={IACIDS},
        year={2024},
        month={3},
        keywords={bidirectional lstm gru convolution accuracy roc-auc score embedding glove},
        doi={10.4108/eai.23-11-2023.2343140}
    }
    
  • Ashish Shinde
    Pranav Shankar
    Atul Atul
    Srikari Rallabandi
    Year: 2024
    Bidirectional LSTM with convolution for toxic comment classification
    IACIDS
    EAI
    DOI: 10.4108/eai.23-11-2023.2343140
Ashish Shinde1,*, Pranav Shankar2, Atul Atul3, Srikari Rallabandi4
  • 1: SRMIST
  • 2: PES University
  • 3: SET
  • 4: Depart of AI, VJIT
*Contact email: ss5504@srmist.edu.in

Abstract

The rapid proliferation of online communication platforms and social media has led to a growing challenge of toxic comments, which encompass harmful, offensive, and inappropriate content that can harm individuals and disrupt online interactions. The accurate detection and filtering of toxic comments are crucial for fostering healthy online discussions and ensuring safe and inclusive social platforms. This paper presents an in-depth exploration of toxic comment classification, with a particular focus on leveraging deep learning techniques. We conducted a comparative analysis of different deep learning architectures, including GRU, Bidirectional GRU, LSTM, Bidirectional LSTM, and a novel model that combines LSTM and convolutional layers. Our study utilized a publicly available dataset of over 106,000 comments categorized into different toxicity classes. Preprocessing and model training were conducted, and the results were evaluated using accuracy and ROC-AUC score metrics. Our findings revealed that the proposed model, which combines LSTM and convolutional layers, outperforms other existing models. It achieved an impressive accuracy of 99.68% and a mean ROC-AUC score of 0.9887. The comprehensive analysis includes a detailed review of related work, model architectures, and extensive experimental results. . Our findings demonstrate the efficacy of the hybrid BiLSTM-CNN model in toxic comment classification.