Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

In-game Toxic Language Detection: Shared Task and Attention Residuals

Yuanzhe Jia, Weixuan Wu, Feiqi Cao, Soyeon Caren Han

TL;DR

This paper tackles the challenge of detecting toxic language in ultra-short in-game chats by formulating a slot-level token tagging task on the CONDA dataset of Dota2 chat logs. It introduces BRAR, a model that combines Bi-LSTM features with attention residuals, label forcing, and a CRF head to improve token-level toxicity labeling. The approach outperforms several strong baselines and demonstrates that incorporating global information via attention residuals and token-label distributions yields better slot predictions, particularly for toxicity-related and domain-specific slots. The work establishes a shared task, provides a robust dataset, and offers design principles for slot-based toxic language detection in real-time gaming environments.

Abstract

In-game toxic language becomes the hot potato in the gaming industry and community. There have been several online game toxicity analysis frameworks and models proposed. However, it is still challenging to detect toxicity due to the nature of in-game chat, which has extremely short length. In this paper, we describe how the in-game toxic language shared task has been established using the real-world in-game chat data. In addition, we propose and introduce the model/framework for toxic language token tagging (slot filling) from the in-game chat. The relevant code is publicly available on GitHub: https://github.com/Yuanzhe-Jia/In-Game-Toxic-Detection

In-game Toxic Language Detection: Shared Task and Attention Residuals

TL;DR

This paper tackles the challenge of detecting toxic language in ultra-short in-game chats by formulating a slot-level token tagging task on the CONDA dataset of Dota2 chat logs. It introduces BRAR, a model that combines Bi-LSTM features with attention residuals, label forcing, and a CRF head to improve token-level toxicity labeling. The approach outperforms several strong baselines and demonstrates that incorporating global information via attention residuals and token-label distributions yields better slot predictions, particularly for toxicity-related and domain-specific slots. The work establishes a shared task, provides a robust dataset, and offers design principles for slot-based toxic language detection in real-time gaming environments.

Abstract

In-game toxic language becomes the hot potato in the gaming industry and community. There have been several online game toxicity analysis frameworks and models proposed. However, it is still challenging to detect toxicity due to the nature of in-game chat, which has extremely short length. In this paper, we describe how the in-game toxic language shared task has been established using the real-world in-game chat data. In addition, we propose and introduce the model/framework for toxic language token tagging (slot filling) from the in-game chat. The relevant code is publicly available on GitHub: https://github.com/Yuanzhe-Jia/In-Game-Toxic-Detection
Paper Structure (6 sections, 3 equations, 1 figure, 2 tables)

This paper contains 6 sections, 3 equations, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Shared Task and Dataset
  3. Methodology
  4. Experiment
  5. Conclusion
  6. Acknowledgments

Figures (1)

  • Figure 1: The overall architecture of BRAR. The dotted arrows represent the proposed Attention Residuals, and the green boxes(LF) present the label-forcing components.