Feature Engineering in the Transformer Era: A Controlled Study on Toxic Comment Classification

Detecting toxic language in user-generated text remains a critical challenge due to linguistic nuance, evolving expressions, and severe class imbalance. While Transformer-based models have established state-of-the-art performance, their significant computational costs pose scalability barriers for real-time moderation. We investigate whether integrating social and contextual metadata—such as user reactions and platform ratings—can bridge the performance gap between computationally efficient classical models and modern deep learning architectures. Using a 40,000-comment subset of the Jigsaw Toxic Comment Classification Challenge, we conduct a controlled, two-phase comparison. We evaluate a Baseline configuration (TF-IDF for classical ensembles vs. raw text for ALBERT) against an Enhanced configuration that fuses text representations with explicit social signals. Our investigation analyzes whether these high-fidelity metadata features allow lightweight models (e.g., LightGBM) to rival the discriminative power of deep Transformers. The findings challenge the prevailing assumption that deep semantic understanding is strictly necessary for high-performance toxicity detection, offering significant implications for the design of scalable, "Green AI" moderation systems.

• Publication year

  2026

• Venue

  Proceedings of the 2026 International Conference on Human-Computer Interaction, Neural Networks and Deep Learning

• Publication date

  2026-01-09

• Fields of study

  Not labeled

• Identifiers
  DOI 10.1145/3795892.3795893
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Social Media Analytics for Disaster Response: Classification and Geospatial Visualization Framework

  2025cited by this paper
• Trade-offs between machine learning and deep learning for mental illness detection on social media

  2025cited by this paper
• More Than a Model: The Compounding Impact of Behavioral Ambiguity and Task Complexity on Hate Speech Detection

  2025cited by this paper
• Beyond words: a hybrid transformer-ensemble approach for detecting hate speech and offensive language on social media

  2025cited by this paper
• Beyond Chat: a Framework for LLMs as Human-Centered Support Systems

  2025cited by this paper
• DamageScope: An Integrated Pipeline for Building Damage Segmentation, Geospatial Mapping, and Interactive Web-Based Visualization

  2025cited by this paper
• An Empirical Comparison of Machine Learning and Deep Learning Models for Automated Fake News Detection

  2025cited by this paper
• Informing Disaster Recovery Through Predictive Relocation Modeling

  2025cited by this paper
• A Comparative Analysis of Deep Learning and Machine Learning Approaches for Spam Identification on Telegram

  2025cited by this paper
• Tutorial on Using Machine Learning and Deep Learning Models for Mental Illness Detection

  2025cited by this paper
• Machine Learning Approaches for Depression Detection on Social Media: A Systematic Review of Biases and Methodological Challenges.

  2024cited by this paper
• A systematic review of machine learning approaches for detecting deceptive activities on social media: methods, challenges, and biases

  2024cited by this paper
• Twitter Hate Speech Detection: A Systematic Review of Methods, Taxonomy Analysis, Challenges, and Opportunities

  2023cited by this paper
• Deep learning for hate speech detection: a comparative study

  2022cited by this paper
• Custodians of the Internet: Platforms, Content Moderation, and the Hidden Decisions that Shape Social Media

  2019cited by this paper
• Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation

  2011cited by this paper
• Adaptive least squares support vector machines filter for hand tremor canceling in microsurgery

  2011cited by this paper
• The relationship between Precision-Recall and ROC curves

  2006cited by this paper
• Greedy function approximation: A gradient boosting machine.

  2001cited by this paper
• Applied Logistic Regression, Second Edition

  1989cited by this paper

• No citing papers are available for this paper.