Project Summary
Posting comments in online discussions has become an important way to ex- ercise one’s right to freedom of expression in the web. This essential right is however under attack as a result of some individuals who hinder respectful reviews with their toxic comments. A toxic comment can be defined as a rude, dis-respectful, or unreasonable comment that is likely to make other users feel bad. It leaves a negative effect on the comment section.
Project Overview: The Toxic Comment Detection project aimed to develop a neural network-based system for identifying and classifying toxic comments in online platforms. Utilizing Python’s machine learning and deep learning libraries, the project focused on leveraging neural networks to enhance moderation efforts, improve user experience, and ensure safe online environments.
- DELIVERABLES
- Utilized Python to develop a neural network model for detecting toxic comments, utilizing deep learning techniques to accurately identify harmful and inappropriate language in text data.
- Collected and preprocessed a Jigsaw and YouTube dataset of online comments, including labeled examples of toxic and non-toxic language, to build a robust training foundation.
- Implemented advanced text processing techniques, such as tokenization, embedding layers, and attention mechanisms, to enhance the model’s ability to understand context and semantics.
- Utilized deep learning frameworks like TensorFlow and Keras to design and train a neural network architecture.
- Applied SMOTE to balance the dataset, in other to achieve a more reliable result.
- Applied WordCloud in visualizing common toxic and non-toxic words.
- Applied feature selection methods, including term frequency-inverse document frequency (TF-IDF) for text cleaning,in other to enhance the model’s accuracy.
- Developed a comprehensive evaluation framework using metrics like precision, recall, F1-score, and ROC-AUC to assess the model’s performance and reliability.
- Achieved a neural network model with 97% accuracy and 7.6% loss.
- ANALYSIS IMPACT
- Successfully developed a neural network-based system for detecting toxic comments, providing an effective tool for moderating online content. The model's integration into live systems will enhance user experience by filtering harmful content, while the detailed evaluation and reporting will offer valuable insights for continuous improvement and deployment strategies.