Research map / Label-efficient and robust learning / Recognition and long-tail
Content Filtering with Inattentive Information Consumers
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.
Models content filtering as a game between a filter and an information consumer who pays rational-inattention costs to examine content, analyzing filter quality improvements under exogenous and endogenous attackers.
Problem
How strategic, cost-incurring consumer attention interacts with content filter quality, and whether improving a filter is unambiguously beneficial when consumers ration attention and attackers respond strategically.
Approach
Content filtering is modeled as a game where the consumer chooses signal fidelity under rational inattention (mutual-information cost) and acts on the filter's forwarded content; the filter's action affects consumer beliefs, unlike information design. Results show filter improvements are weakly Pareto-improving under aligned utilities, can be inefficient when the filter ignores consumer informatio
Key results
- With an exogenous attacker and aligned utilities, improving filter quality is weakly Pareto-improving, with zero marginal value until the filter is sufficiently accurate (Theorem 4.2).
- With semi-aligned utilities, all equilibria may be Pareto-inefficient for low-quality filters, and this inefficiency vanishes once the filter is sufficiently accurate (Theorems 5.1, 5.2).
- With a strategic attacker, the consumer incurs no information cost in any equilibrium (Theorem 6.1), and sufficiently improving the filter can strictly decrease equilibrium payoffs for both filter and consumer when pi0/pi1 < Lambda (Theorem 6.2).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Improve filter quality with an exogenous attacker and aligned utilities when the differentiating profile is optimal and qdif < qL | Content filtering game with aligned utilities · Exogenous attacker | equilibrium payoff: ∂V*/∂π0 = q/(1 − q) c2 > 0; ∂V*/∂π1 = −(c1 + b) < 0 | Helped |
| Improve filter quality with an exogenous attacker and aligned utilities when qdif > qH or forwarding is better than differentiating | Content filtering game with aligned utilities · Exogenous attacker | equilibrium payoff: 0 | No clear effect |
| Improve filter quality with an exogenous attacker and aligned utilities when the differentiating profile is optimal and qdif is between qL and qH | Content filtering game with aligned utilities · Exogenous attacker | equilibrium payoff: ∂V*/∂π0 = q/(1 − q) λ · log(qH/qdif) > 0; ∂V*/∂π1 = λ · log((1 − qH)/(1 − qdif)) < 0 | Helped |
| Improve filter quality past the threshold Q = Λ | Content filtering game with semi-aligned utilities · Exogenous attacker; qL < qdif < qH < q | Pareto efficiency | Helped |
| Further improve filter quality while Q > Λ under the differentiating equilibrium | Content filtering game with semi-aligned utilities · Exogenous attacker; qL < qdif < qH < q | equilibrium payoff | Helped |
| Improve filter quality sufficiently when the attacker is strategic and π0/π1 < Λ | Content filtering game with strategic attacker and aligned utilities · Endogenous attacker | equilibrium utility | Hurt |
Surprising findings
- With a strategic attacker, improving the filter can make both the filter and the consumer worse off, as the attacker raises its attack propensity.
- Under semi-aligned utilities, a low-quality filter cannot credibly commit to differentiating and the resulting equilibria are Pareto-inefficient.
Limitations
- Single-round stationary game with homogeneous, non-adaptive players; dynamics are not modeled.
- Binary legitimacy, filter signal, action and consumer signal; richer signals are left for future work.
- Information cost is the specific rational-inattention mutual-information form.
Experimental setup
Metrics: equilibrium payoffs value of technological change
Baselines: information design with rational inattention (Matyskova and Montes 2023)
Builds on: rational inattention (Sims 2003) Papanastasiou 2020
Closest papers by meaning
- Beyond Self-Interest: How Group Strategies Reshape Content Creation in Recommendation Platforms? (ICML 2025 · similarity 0.84)
- Private Blotto: Viewpoint Competition with Polarized Agents (AAAI 2025 · similarity 0.83)
- On the Power of Strategic Corpus Enrichment in Content Creation Games (AAAI 2025 · similarity 0.82)
- Recommender Ecosystems: A Mechanism Design Perspective on Holistic Modeling and Optimization (AAAI 2024 · similarity 0.82)
- Policy Design for Two-sided Platforms with Participation Dynamics (ICML 2025 · similarity 0.82)
- The Burden of Interactive Alignment with Inconsistent Preferences (NeurIPS 2025 · similarity 0.81)
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.