Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MGDIL: Multi-Granularity Summarization and Domain-Invariant Learning for Cross-Domain Social Bot Detection

Boyu Qiao, Yunman Chen, Kun Li, Wei Zhou, Songlin Hu, Yunya Song

Abstract

Social bots increasingly infiltrate online platforms through sophisticated disguises, threatening healthy information ecosystems. Existing detection methods often rely on modality specific cues or local contextual features, making them brittle when modalities are missing or inputs are incomplete. Moreover, most approaches assume similar train test distributions, which limits their robustness to out of distribution (OOD) samples and emerging bot types. To address these challenges, we propose Multi Granularity Summarization and Domain Invariant Learning (MGDIL), a unified framework for robust social bot detection under domain shift. MGDIL first transforms heterogeneous signals into unified textual representations through LLM based multi granularity summarization. Building on these representations, we design a collaborative optimization framework that integrates task oriented LLM instruction tuning with domain invariant representation learning. Specifically, task oriented instruction tuning enhances the LLMs ability to capture subtle semantic cues and implicit camouflage patterns, while domain adversarial learning and cross domain contrastive learning are jointly employed to mitigate distribution shifts across datasets and time periods. Through this joint optimization, MGDIL learns stable and discriminative domain invariant features, improving cross domain social bot detection through better distribution alignment, stronger intra class compactness, and clearer inter class separation.

MGDIL: Multi-Granularity Summarization and Domain-Invariant Learning for Cross-Domain Social Bot Detection

Abstract

Social bots increasingly infiltrate online platforms through sophisticated disguises, threatening healthy information ecosystems. Existing detection methods often rely on modality specific cues or local contextual features, making them brittle when modalities are missing or inputs are incomplete. Moreover, most approaches assume similar train test distributions, which limits their robustness to out of distribution (OOD) samples and emerging bot types. To address these challenges, we propose Multi Granularity Summarization and Domain Invariant Learning (MGDIL), a unified framework for robust social bot detection under domain shift. MGDIL first transforms heterogeneous signals into unified textual representations through LLM based multi granularity summarization. Building on these representations, we design a collaborative optimization framework that integrates task oriented LLM instruction tuning with domain invariant representation learning. Specifically, task oriented instruction tuning enhances the LLMs ability to capture subtle semantic cues and implicit camouflage patterns, while domain adversarial learning and cross domain contrastive learning are jointly employed to mitigate distribution shifts across datasets and time periods. Through this joint optimization, MGDIL learns stable and discriminative domain invariant features, improving cross domain social bot detection through better distribution alignment, stronger intra class compactness, and clearer inter class separation.

Paper Structure

This paper contains 29 sections, 26 equations, 11 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Conventional Social Bot Detection.
  4. OOD Social Bot Detection
  5. Methodology
  6. Multi-Granularity Summarization
  7. Domain-Invariant Representation Learning
  8. Relation-aware GNN for Structural Enhancement
  9. Experiments
  10. Experiment Setup
  11. Main Results
  12. Multi-Granularity Summarization Study
  13. Domain-Invariant Learning Study
  14. Heterogeneous Graph-based Study
  15. Hyperparameter Analysis
  16. ...and 14 more sections

Figures (11)

  • Figure 1: Two Core Challenges in Social Bot Detection.
  • Figure 2: Overview of the MGDIL framework. MGDIL first summarizes user information into unified structured textual inputs, and then performs domain-invariant learning, and graph-based representation learning for robust social bot detection.
  • Figure 3: Comparison Between Raw Historical Posts and Summarized Historical Posts.
  • Figure 4: Hyperparameter analysis of $\lambda_{adv}$ and $\lambda_{con}$, where one loss weight is varied while the other is fixed at 0.2.
  • Figure 5: Prompt Template for LLM-based Multi-Dimensional Summarization
  • ...and 6 more figures