Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ToxVidLM: A Multimodal Framework for Toxicity Detection in Code-Mixed Videos

Krishanu Maity, A. S. Poornash, Sriparna Saha, Pushpak Bhattacharyya

TL;DR

The paper tackles toxicity detection in code-mixed Hindi–English videos by introducing ToxCMM, a dataset of 931 videos with 4021 sentence-level utterances annotated for toxicity, sentiment, and severity. It presents ToxVidLM, a multimodal multitask framework with an Encoder, Cross-Modal Synchronization, and Multitask modules to fuse text, audio, and video signals for improved toxicity detection and sentiment/severity analysis. Experiments show that text encodings dominate unimodal performance, while integrating audio and video via a text-guided synchronization and gated fusion approach yields state-of-the-art results on all three tasks, with a multitask setup providing additional gains. The dataset and framework collectively advance safe online content moderation for low-resource, code-mixed languages and open avenues for context-aware, scalable toxicity analysis in multimedia content.

Abstract

In an era of rapidly evolving internet technology, the surge in multimodal content, including videos, has expanded the horizons of online communication. However, the detection of toxic content in this diverse landscape, particularly in low-resource code-mixed languages, remains a critical challenge. While substantial research has addressed toxic content detection in textual data, the realm of video content, especially in non-English languages, has been relatively underexplored. This paper addresses this research gap by introducing a benchmark dataset, the first of its kind, consisting of 931 videos with 4021 code-mixed Hindi-English utterances collected from YouTube. Each utterance within this dataset has been meticulously annotated for toxicity, severity, and sentiment labels. We have developed an advanced Multimodal Multitask framework built for Toxicity detection in Video Content by leveraging Language Models (LMs), crafted for the primary objective along with the additional tasks of conducting sentiment and severity analysis. ToxVidLM incorporates three key modules - the Encoder module, Cross-Modal Synchronization module, and Multitask module - crafting a generic multimodal LM customized for intricate video classification tasks. Our experiments reveal that incorporating multiple modalities from the videos substantially enhances the performance of toxic content detection by achieving an Accuracy and Weighted F1 score of 94.29% and 94.35%, respectively.

ToxVidLM: A Multimodal Framework for Toxicity Detection in Code-Mixed Videos

TL;DR

The paper tackles toxicity detection in code-mixed Hindi–English videos by introducing ToxCMM, a dataset of 931 videos with 4021 sentence-level utterances annotated for toxicity, sentiment, and severity. It presents ToxVidLM, a multimodal multitask framework with an Encoder, Cross-Modal Synchronization, and Multitask modules to fuse text, audio, and video signals for improved toxicity detection and sentiment/severity analysis. Experiments show that text encodings dominate unimodal performance, while integrating audio and video via a text-guided synchronization and gated fusion approach yields state-of-the-art results on all three tasks, with a multitask setup providing additional gains. The dataset and framework collectively advance safe online content moderation for low-resource, code-mixed languages and open avenues for context-aware, scalable toxicity analysis in multimedia content.

Abstract

In an era of rapidly evolving internet technology, the surge in multimodal content, including videos, has expanded the horizons of online communication. However, the detection of toxic content in this diverse landscape, particularly in low-resource code-mixed languages, remains a critical challenge. While substantial research has addressed toxic content detection in textual data, the realm of video content, especially in non-English languages, has been relatively underexplored. This paper addresses this research gap by introducing a benchmark dataset, the first of its kind, consisting of 931 videos with 4021 code-mixed Hindi-English utterances collected from YouTube. Each utterance within this dataset has been meticulously annotated for toxicity, severity, and sentiment labels. We have developed an advanced Multimodal Multitask framework built for Toxicity detection in Video Content by leveraging Language Models (LMs), crafted for the primary objective along with the additional tasks of conducting sentiment and severity analysis. ToxVidLM incorporates three key modules - the Encoder module, Cross-Modal Synchronization module, and Multitask module - crafting a generic multimodal LM customized for intricate video classification tasks. Our experiments reveal that incorporating multiple modalities from the videos substantially enhances the performance of toxic content detection by achieving an Accuracy and Weighted F1 score of 94.29% and 94.35%, respectively.
Paper Structure (17 sections, 6 equations, 4 figures, 4 tables)

This paper contains 17 sections, 6 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Toxic Code-Mixed Multimodal ( ToxCMM) Dataset Creation
  4. Data Annotation
  5. Dataset Statistics
  6. Methodology
  7. Encoder Module
  8. Cross Modal Synchronization Module
  9. Multitask Module
  10. Loss Function
  11. Experimental Results and Analysis
  12. Findings from Experiments
  13. Ablation Study
  14. Conclusion and Future Works
  15. Limitations
  16. ...and 2 more sections

Figures (4)

  • Figure 1: Some samples from annotated ToxCMM dataset; The yellow highlighted words are in English.
  • Figure 2: Architecture of proposed ToxVidLM model.
  • Figure 3: A schematic of baselines models as mentioned in das2023hatemm
  • Figure 4: Human annotation versus model’s prediction for qualitative analysis with different settings