Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AggregHate: An Efficient Aggregative Approach for the Detection of Hatemongers on Social Platforms

Tom Marzea, Abraham Israeli, Oren Tsur

TL;DR

This paper tackles hate-monger detection on social platforms by moving from post-centric detection to a user-centric, multimodal framework that fuses utterance-level predictions with social-context signals. It presents a formal aggregation hierarchy, including naive, relational, distributional, and multimodal schemes, and benchmarks these against diffusion- and GNN-based baselines. Evaluations on Echo (Twitter), Gab, and Parler show that incorporating network context and distributional cues yields substantial improvements over text-only methods, with platform-specific patterns in the effectiveness of relational versus distributional components. The work also contributes a Parler user-annotated dataset and discusses practical considerations for deployment, including ethical safeguards and the importance of human-in-the-loop interventions for high-stakes moderation.

Abstract

Automatic detection of online hate speech serves as a crucial step in the detoxification of the online discourse. Moreover, accurate classification can promote a better understanding of the proliferation of hate as a social phenomenon. While most prior work focus on the detection of hateful utterances, we argue that focusing on the user level is as important, albeit challenging. In this paper we consider a multimodal aggregative approach for the detection of hate-mongers, taking into account the potentially hateful texts, user activity, and the user network. We evaluate our methods on three unique datasets X (Twitter), Gab, and Parler showing that a processing a user's texts in her social context significantly improves the detection of hate mongers, compared to previously used text and graph-based methods. Our method can be then used to improve the classification of coded messages, dog-whistling, and racial gas-lighting, as well as inform intervention measures. Moreover, our approach is highly efficient even for very large datasets and networks.

AggregHate: An Efficient Aggregative Approach for the Detection of Hatemongers on Social Platforms

TL;DR

This paper tackles hate-monger detection on social platforms by moving from post-centric detection to a user-centric, multimodal framework that fuses utterance-level predictions with social-context signals. It presents a formal aggregation hierarchy, including naive, relational, distributional, and multimodal schemes, and benchmarks these against diffusion- and GNN-based baselines. Evaluations on Echo (Twitter), Gab, and Parler show that incorporating network context and distributional cues yields substantial improvements over text-only methods, with platform-specific patterns in the effectiveness of relational versus distributional components. The work also contributes a Parler user-annotated dataset and discusses practical considerations for deployment, including ethical safeguards and the importance of human-in-the-loop interventions for high-stakes moderation.

Abstract

Automatic detection of online hate speech serves as a crucial step in the detoxification of the online discourse. Moreover, accurate classification can promote a better understanding of the proliferation of hate as a social phenomenon. While most prior work focus on the detection of hateful utterances, we argue that focusing on the user level is as important, albeit challenging. In this paper we consider a multimodal aggregative approach for the detection of hate-mongers, taking into account the potentially hateful texts, user activity, and the user network. We evaluate our methods on three unique datasets X (Twitter), Gab, and Parler showing that a processing a user's texts in her social context significantly improves the detection of hate mongers, compared to previously used text and graph-based methods. Our method can be then used to improve the classification of coded messages, dog-whistling, and racial gas-lighting, as well as inform intervention measures. Moreover, our approach is highly efficient even for very large datasets and networks.
Paper Structure (34 sections, 6 equations, 1 figure, 7 tables)

This paper contains 34 sections, 6 equations, 1 figure, 7 tables.

Table of Contents

  1. Introduction
  2. Contribution
  3. Related Work
  4. Multimodal Aggregative Approaches
  5. Aggregative Approaches
  6. Utterance-level Classification ($C^T$)
  7. User-level Classification ($C^U$)
  8. Naive Aggregation with Fixed Threshold
  9. Relational Aggregation
  10. Distributional Aggregation
  11. Multimodal Aggregation
  12. Social-aware Baselines
  13. DeGroot's Diffusion
  14. Graph Neural Networks (GNNs)
  15. Datasets and Annotation
  16. ...and 19 more sections

Figures (1)

  • Figure 1: The Density of the mean utterance scores per user for each Dataset