Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DeHate: A Stable Diffusion-based Multimodal Approach to Mitigate Hate Speech in Images

Dwip Dalal, Gautam Vashishtha, Anku Rani, Aishwarya Reganti, Parth Patwa, Mohd Sarique, Chandan Gupta, Keshav Nath, Viswanatha Reddy, Vinija Jain, Aman Chadha, Amitava Das, Amit Sheth, Asif Ekbal

TL;DR

This work tackles hate speech in multimodal content by creating the DeHate dataset and a diffusion-based dehatification pipeline. It introduces DeHater, a CLIP-guided, FiLM-conditioned model that generates masked outputs to blur hateful regions localized by Diffusion Attentive Attribution Maps (DAAM). The dataset and shared task use an IoU-based evaluation, with top-performing systems achieving approximately 0.55 IOU, illustrating the task's difficulty. The approach advances ethical AI for social media by enabling targeted dehatification and establishing a benchmark for multimodal hate mitigation, with future work including LLM-based justification and multilingual extension.

Abstract

The rise in harmful online content not only distorts public discourse but also poses significant challenges to maintaining a healthy digital environment. In response to this, we introduce a multimodal dataset uniquely crafted for identifying hate in digital content. Central to our methodology is the innovative application of watermarked, stability-enhanced, stable diffusion techniques combined with the Digital Attention Analysis Module (DAAM). This combination is instrumental in pinpointing the hateful elements within images, thereby generating detailed hate attention maps, which are used to blur these regions from the image, thereby removing the hateful sections of the image. We release this data set as a part of the dehate shared task. This paper also describes the details of the shared task. Furthermore, we present DeHater, a vision-language model designed for multimodal dehatification tasks. Our approach sets a new standard in AI-driven image hate detection given textual prompts, contributing to the development of more ethical AI applications in social media.

DeHate: A Stable Diffusion-based Multimodal Approach to Mitigate Hate Speech in Images

TL;DR

This work tackles hate speech in multimodal content by creating the DeHate dataset and a diffusion-based dehatification pipeline. It introduces DeHater, a CLIP-guided, FiLM-conditioned model that generates masked outputs to blur hateful regions localized by Diffusion Attentive Attribution Maps (DAAM). The dataset and shared task use an IoU-based evaluation, with top-performing systems achieving approximately 0.55 IOU, illustrating the task's difficulty. The approach advances ethical AI for social media by enabling targeted dehatification and establishing a benchmark for multimodal hate mitigation, with future work including LLM-based justification and multilingual extension.

Abstract

The rise in harmful online content not only distorts public discourse but also poses significant challenges to maintaining a healthy digital environment. In response to this, we introduce a multimodal dataset uniquely crafted for identifying hate in digital content. Central to our methodology is the innovative application of watermarked, stability-enhanced, stable diffusion techniques combined with the Digital Attention Analysis Module (DAAM). This combination is instrumental in pinpointing the hateful elements within images, thereby generating detailed hate attention maps, which are used to blur these regions from the image, thereby removing the hateful sections of the image. We release this data set as a part of the dehate shared task. This paper also describes the details of the shared task. Furthermore, we present DeHater, a vision-language model designed for multimodal dehatification tasks. Our approach sets a new standard in AI-driven image hate detection given textual prompts, contributing to the development of more ethical AI applications in social media.

Paper Structure

This paper contains 11 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset Curation
  4. Hatenorm Dataset
  5. DeHate Dataset
  6. Shared Task details
  7. Evaluation
  8. Participating systems
  9. Methodology
  10. Results
  11. Conclusion and Future Work

Figures (3)

  • Figure 1: A glimpse on the dataset that we curated. The first image is the image generated from the tweet, the second image is dehated version and the third image is the output of the method that we propose for dehatification task.
  • Figure 2: Stable diffusion output with attention maps.
  • Figure 3: Architecture diagram of our proposed system.