Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{191772,
        author = {Nelabanda Indu lahari and Mangali Mamatha and M.Saniya Tabassum and kummari Nikhitha and Dr C V Madhusudhan Reddy and Chakali Rambabu},
        title = {Toxic Comment Classification for Social Media Using IBM Services},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {8},
        pages = {8760-8765},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=191772},
        abstract = {The ToxiShield is an original machine learning solution, developed for the purpose of improving online communication quality by effectively identifying toxic messages in six classes of toxicity, namely toxic, severe toxic, obscene, threat, insult, and identity hate, based on the Jigsaw Toxic Comment Classification Challenge dataset available at Kaggle, which includes 159,571 labeled toxic and non-toxic Wikipedia messages. The model uses extensive preprocessing in NLTK for cleaning messages, TF-IDF and Count Vectorization for finding features, and state-of-the-art dimensionality reduction techniques using Truncated SVD and PCA, and multi-label classification using SGDRegreesor, DecisionTreeRegreesor, and deep learning techniques utilizing TensorFlow and Keras libraries. Using scikit-learn libraries for effective and extensive use of evaluation measures such as ROC-AUC, this solution removes imbalances in toxicity levels using strategic oversampling and plots using Matplotlib, Seaborn, and Word Cloud libraries. Scaled as an efficient and high-quality production-level API solution, ToxiShield provides actual-time toxicity probability results, which are more effective than existing approaches and have significant gaps in high-quality content moderation solutions for SNSs.},
        keywords = {Toxic comment classification, multi-label toxicity detection, TF-IDF vectorization, BERT fine-tuning, IBM Cloud deployment, NLP preprocessing, ROC-AUC evaluation, social media safety},
        month = {January},
        }