5. ■ Study 1: A Survey on Toxic Comment Classification
• Authors: Jiawei Zhang, Xiaofei Sun, and Bing Liu
• Key Findings:
o Presents a comprehensive overview of toxic comment classification
techniques
o Highlights various machine learning algorithms and datasets used for
classification
■ Study 2: Deep Learning for Toxic Comment Classification in Social
Media
• Authors: Mohamed Abdelwahab, Amr Elsheikh, and Hany Farid
• Key Findings:
o Utilizes convolutional neural networks (CNNs) to extract informative
features from comment text
o Achieves remarkable accuracy (96.2%) on an English-language dataset
6. ■ Study 3: Interpretable Multi-Labeled Toxic Comment Classification in
Bengali Using Deep Learning
• Authors: Belal, T. A., Shahariar, G. M., and Kabir, M. H.
• Key Findings:
o Introduces an interpretable multi-labeled approach for toxic comment
classification in Bengali
o Combines LSTM and CNN models to achieve accuracy rates of 89.42% for binary
classification and 78.92% for multi-label classification
■ Study 4: Toxic Comment Classification in Urdu: A Novel Approach Based on
Word Embeddings and Transformer Models
• Authors: Muhammad Usman, Muhammad Saad, and Muhammad
Ahsan
• Key Findings:
o Employs word embeddings and a transformer model for classification
o Achieves an impressive accuracy of 95.7% on an Urdu-language dataset
7. CONS.
■ All the General and Conference Paper we read, They have same disadvantage in it.
They have only work with one category. Means
there result will tell you that the comment is Toxic
or Not
9. Solution
■ Categorized the data in to 6 sub groups
Toxic Severe Toxic Obscene Insult Identity Hate Healthy
■ Applied several Deep and Machine Learning Models
CNN BGRU BLSTM SVM Log Reg NB SVM
■ But in our case the best performing model is SVM
21. ■ Toxic comment classification is crucial for fostering a safer
and more inclusive online environment.
■ We have explored the challenges associated with toxic
comment classification and proposed a novel methodology
for multiclassification using distinct languages, Roman
Urdu, and English.
■ Our findings emphasize the significance of algorithm
selection, data refinement, and feature dimensionality in
achieving accurate classification.
22. ■ We conducted a comparative analysis of various machine learning
algorithms, including SVM, Log Reg, and NB, to assess their effectiveness in
identifying toxic comments.
■ Our research highlights the importance of tailored approaches in
multilingual environments, particularly considering cultural nuances.
■ We believe our findings can contribute to reducing the impact of toxic
content online and promoting a positive e-environment.
■ The study suggests future directions for toxic comment classification,
including ensemble methods for improved performance and transformer
models for enhanced accuracy and robustness.
■ As transformer models evolve, they hold promise for handling larger
datasets and targeting more languages, making them a valuable tool for
multilingual toxic comment classification.
Editor's Notes
In this sense, "plagued" can be used to describe a situation that is causing someone or something a lot of difficulty