Research map / Learning theory and algorithms / Submodular and game algorithms
A Persuasive Approach to Combating Misinformation
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 noisy Bayesian persuasion, where a platform signals predicted popularity/misinformation of a user's draft post, characterizes optimal signaling as an LP, and analyzes a stateful performative dynamic with synthetic-data experiments.
Problem
Social media platforms want to reduce misinformation sharing without censorship, but their predictions of post popularity and misinformation are imperfect, and the user's future behaviour responds to interventions.
Approach
Model platform-user interaction as a Bayesian Stackelberg game where the platform signals based on noisy classifier predictions; the optimal scheme is an LP over effective signaling constrained by the confusion matrix, with a performative stateful update of the user's content prior.
Key results
- Worked example: platform utility rises from 0.45 to 0.64 and misinformation share falls from 30% to 17% with 90% accurate classifiers, with user utility unchanged at 0.
- Experiments on synthetic instances show about 10% reduction in misinformation shared even at high classifier error, rising to around 20% when classifier errors are below 0.15.
- Theory: optimal platform utility is non-decreasing under confusion-matrix change iff row convex hull inclusion holds, and is Lipschitz in the confusion matrix; the performative process with lambda=0 converges in 2(|Theta|-1) iterations to a stable share-recommending point.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply noisy persuasion using 90% accurate classifiers | The synthetic example · Synthetic example | Platform utility: 0.45 -> 0.64 | Helped |
| Apply noisy persuasion using 90% accurate classifiers | The synthetic example · Synthetic example | % of shared posts that are misinformation: 30 -> 17 | Helped |
| Apply noisy persuasion using 90% accurate classifiers | The synthetic example · Synthetic example | User utility: 0 -> 0 | No clear effect |
| Apply noisy persuasion when both classifiers perform poorly | The synthetic setting · Synthetic instances, single application | Misinformation reduction in shared posts: 10% | Helped |
| Use classifiers with errors below 0.15 when applying noisy persuasion | The synthetic setting · Synthetic instances, single application | Misinformation reduction in shared posts: around 20% | Helped |
| Apply noisy persuasion over time until the stable point | The performative persuasion process · Synthetic instances, stable point | Misinformation shared | Helped |
Surprising findings
- Results are more sensitive to the misinformation classifier's accuracy than to the validation classifier's accuracy.
Limitations
- Experiments use synthetic data rather than real user or platform data.
- Assumes a single perfectly rational user with the platform knowing user utility.
- Lacks necessary conditions guaranteeing increased platform utility at the stable point (only a sufficient condition given).
Experimental setup
Datasets: synthetic instances (random priors and utilities)
Benchmarks: synthetic persuasion instances
Metrics: platform utility percentage of shared posts that are misinformation user utility
Baselines: no persuasion (prior)
Builds on: Bayesian persuasion performative prediction Kamenica and Gentzkow concave closure
Closest papers by meaning
- Persuasive Prediction via Decision Calibration (NeurIPS 2026 · similarity 0.87)
- Online Bayesian Persuasion Without a Clue (NeurIPS 2024 · similarity 0.84)
- Beyond Detection: Exploring Evidence-based Multi-Agent Debate for Misinformation Intervention and Persuasion (AAAI 2026 · similarity 0.83)
- Markov Persuasion Processes: Learning to Persuade From Scratch (NeurIPS 2025 · similarity 0.83)
- Bayesian Persuasion with Externalities: Exploiting Agent Types (AAAI 2025 · similarity 0.82)
- Learning to Persuade a Biased Receiver (NeurIPS 2026 · similarity 0.82)
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.