Reproduce “ImpliHateVid: A Benchmark Dataset and Two-stage Contrastive Learning Framework for Implicit Hate Speech Detection in Videos” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mohammad Zia Ur Rehman, Anukriti Bhatnagar, Omkar Kabde, Shubhi Bansal, Nagendra Kumar
Researchers from the Indian Institute of Technology Indore and Chaitanya Bharathi Institute of Technology introduced ImpliHateVid, a benchmark dataset for implicit hate speech in videos, alongside a two-stage contrastive learning framework. The framework improved hate speech detection, achieving an 87.73% F1-score on ImpliHateVid for binary classification and a 69.18% macro-F1 for multiclass classification on the same dataset.
Key insights
- Detecting implicit hate speech in videos necessitates a multimodal approach that effectively combines textual, visual, and audio cues, as unimodal methods perform considerably worse.
- A two-stage contrastive learning strategy helps to effectively align and integrate features from various modalities, thereby enhancing the model's capability to discern subtle hateful content.
- Complementary features such as emotion, sentiment, and image captions provide essential additional context, which further improves the detection of implicit hate speech by capturing subtle cues.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.