Designed to Spread: Generative Approaches to Enhance Information Diffusion

TL;DR

The paper introduces Diffusion-Oriented Content Generation (DOCG), a framework that automatically rewrites content to maximize diffusion within a target audience without relying on explicit network topology. It couples an influence indicator, based on a pairwise diffusion estimator, with an information editor that uses reinforcement learning and modality-specific actions to generate semantically faithful, audience-tailored text or image revisions; the editor optimizes a reward that balances diffusion gain with semantic fidelity. Empirical results on real-world Twitter datasets show strong predictive accuracy for the influence indicator and significant diffusion gains for both textual and visual content, supported by a user study indicating favorable sharing propensity. The work advances content-level diffusion control, enabling scalable, audience-aware dissemination and offering avenues for multimodal extensions, though it acknowledges latency and noisy diffusion signals as practical challenges.

Abstract

Social media has fundamentally transformed how people access information and form social connections, with content expression playing a critical role in driving information diffusion. While prior research has focused largely on network structures and tipping point identification, it provides limited tools for automatically generating content tailored for virality within a specific audience. To fill this gap, we propose the novel task of DOCG and introduce an information enhancement algorithm for generating content optimized for diffusion. Our method includes an influence indicator that enables content-level diffusion assessment without requiring access to network topology, and an information editor that employs reinforcement learning to explore interpretable editing strategies. The editor leverages generative models to produce semantically faithful, audience-aware textual or visual content. Experiments on real-world social media datasets and user study demonstrate that our approach significantly improves diffusion effectiveness while preserving the core semantics of the original content.

Figures (5)

• Figure 1: Overview of our proposed information enhancement framework, consisting of two main components: the influence indicator, and the information editor.
• Figure 2: Our information editor for editing message content $c$. The policy network generates editing actions to iteratively revise the content, guided by the reward signal.
• Figure 3: Distribution of influence scores for messages on observed (orange) and permuted (blue) combinations. *** denotes statistical significance at $p < 0.001$.
• Figure 4: Representative text and image examples across different levels of influence scores $\mathcal{L}_U(c)$.
• Figure 5: Qualitative comparison with baselines. Each piece of content is annotated with its corresponding influence score.