Research map / Probabilistic and scientific modeling / Molecules and proteins
Adversarial Socialbots Modeling Based on Structural Information Principles
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes SIASM, a structural-information-principles framework that uses optimal encoding trees, an entropy-based influence score for pruning, and conditional structural entropy with PPO to guide socialbot follower selection under a robust detector.
Problem
Adversarial socialbot modeling is incomplete: existing hierarchical RL approaches learn activity types from scratch without global network structure and compute follower selection over all users' local features, which is inefficient; this hinders proactive bot detection.
Approach
Uses high-dimensional structural entropy minimization to build a hierarchical community tree of the social graph, scores community influence via assigned structural entropy to prune uninfluential users, and defines conditional structural entropy between the socialbot and each user to initialize follower selection, refined by PPO.
Key results
- On homogeneous data, SIASM achieves average episode reward 831.19 vs. 771.34 for ACORN (+16.32% on synthetic reward) and episode length 108.45 vs. 97.12 on synthetic graphs.
- On heterogeneous real-life graphs, SIASM achieves episode reward 840.89 and length 80.0, versus ACORN's 821.83 and 71.73.
- In survival tests, SIASM persists about 2.4K timesteps with network influence ratio 0.99 across real-life scenarios.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SIASM instead of ACRON | Homogeneous social networks · Homogeneous average performance | episode reward: +5.66(0.69%) | Helped |
| Use SIASM instead of ACRON | Homogeneous social networks · Homogeneous average performance | episode length: +12.27(14.87%) | Helped |
| Use SIASM instead of SIASM-H | Heterogeneous social networks · Heterogeneous average performance | episode reward: +2.92(0.35%) | Helped |
| Use SIASM instead of ACRON | Heterogeneous social networks · Heterogeneous average performance | episode length: +5.09(5.64%) | Helped |
| Use SIASM instead of ACRON | Homogeneous social networks · Homogeneous real-life, p=0.25 | network influence ratio | Helped |
| Use SIASM instead of ACRON | Homogeneous social networks · Homogeneous real-life, p=0.25 | survival steps | Helped |
| Remove user filtration (ablation) | SIASM · Homogeneous social networks | policy-learning quality | Hurt |
| Add user filtration (ablation) | SIASM · Homogeneous social networks | policy-learning quality | Helped |
| Remove follower selection (ablation) | SIASM · Homogeneous social networks | policy-learning quality | Hurt |
| Add follower selection (ablation) | SIASM · Homogeneous social networks | policy-learning quality | Helped |
Surprising findings
- Real-life homogeneous reward gains over ACORN are small (about 0.37%), suggesting most gains come from synthetic graphs.
Limitations
- Evaluated against a robust detector with 90% accuracy only in the framed setting; detector is simulated.
- Single socialbot modeled; multi-bot adversarial settings left for future work.
- Real-life homogeneous gains over ACORN are marginal.
Experimental setup
Datasets: Twitter trending articles on US presidential election and COVID-19 (1500 users) Higgs Twitter Dataset synthetic stochastic networks
Benchmarks: Twitter propagation networks (homogeneous) Higgs Twitter Dataset (heterogeneous)
Metrics: episode reward (network influence) episode length (sustainable stealthiness) network influence ratio survival timesteps
Baselines: ACORN ACORN-H SIASM-H CELF DEGREE
Builds on: ACORN HCSE R-GCN PPO Independence Cascade Model
Closest papers by meaning
- Enhancing Multi-Scale Diffusion Prediction via Sequential Hypergraphs and Adversarial Learning (AAAI 2024 · similarity 0.86)
- Enhancing Robustness of Graph Neural Networks on Social Media with Explainable Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
- TSGAN: Temporal Social Graph Attention Network for Aggressive Behavior Forecasting (AAAI 2025 · similarity 0.85)
- DiffIM: Differentiable Influence Minimization with Surrogate Modeling and Continuous Relaxation (AAAI 2025 · similarity 0.83)
- Spear and Shield: Adversarial Attacks and Defense Methods for Model-Based Link Prediction on Continuous-Time Dynamic Graphs (AAAI 2024 · similarity 0.83)
- Non-Euclidean Mixture Model for Social Network Embedding (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.