DREAMS: A Social Exchange Theory-Informed Modeling of Misinformation Engagement on Social Media

TL;DR

DREAMS reframes misinformation engagement on social media as a dynamic, platform-conditioned social exchange process and couples Social Exchange Theory with a modular, disentangled sequence model. The architecture combines disentangled representations, latent belief states, FiLM-based platform adaptation, and a dual-timescale memory system to capture both short-term reciprocity and long-term exchange dynamics across seven platforms. On a large cross-platform dataset (2.36M posts across 7 platforms, 2021–2025), DREAMS achieves state-of-the-art engagement prediction (MAPE ≈ 19.25%), substantially outperforming strong baselines and providing interpretable insights into platform-specific exchange rates and emotional currency. The work demonstrates that grounding neural models in behavioral theory enhances predictive performance and cross-platform generalization, with implications for understanding and mitigating misinformation spread in evolving online ecosystems.

Abstract

Social media engagement prediction is a central challenge in computational social science, particularly for understanding how users interact with misinformation. Existing approaches often treat engagement as a homogeneous time-series signal, overlooking the heterogeneous social mechanisms and platform designs that shape how misinformation spreads. In this work, we ask: ``Can neural architectures discover social exchange principles from behavioral data alone?'' We introduce \textsc{Dreams} (\underline{D}isentangled \underline{R}epresentations and \underline{E}pisodic \underline{A}daptive \underline{M}odeling for \underline{S}ocial media misinformation engagements), a social exchange theory-guided framework that models misinformation engagement as a dynamic process of social exchange. Rather than treating engagement as a static outcome, \textsc{Dreams} models it as a sequence-to-sequence adaptation problem, where each action reflects an evolving negotiation between user effort and social reward conditioned by platform context. It integrates adaptive mechanisms to learn how emotional and contextual signals propagate through time and across platforms. On a cross-platform dataset spanning $7$ platforms and 2.37M posts collected between 2021 and 2025, \textsc{Dreams} achieves state-of-the-art performance in predicting misinformation engagements, reaching a mean absolute percentage error of $19.25$\%. This is a $43.6$\% improvement over the strongest baseline. Beyond predictive gains, the model reveals consistent cross-platform patterns that align with social exchange principles, suggesting that integrating behavioral theory can enhance empirical modeling of online misinformation engagement. The source code is available at: https://github.com/ltian678/DREAMS.

Figures (8)

• Figure 1: Overview of social media misinformation engagement dynamics and the proposed engagement pyramid.
• Figure 2: Overview of the Dreams architecture for cross-platform misinformation engagement prediction. (Top) The encoder disentangles input features into three latent factors (content ($z_c$), platform ($z_p$), temporal dynamics ($z_t$)), which then projected onto a hypersphere for scale-invariant representations. (Bottom) The aligned representations feed into four SET-driven components: (i) a Latent Belief State to track exchange context, (ii) Dynamic Architecture Adaptation via FiLM layers with sparse gating, (iii) a Dual-Timescale Memory System with fast and slow adapters, and (iv) a Replay Buffer to retain past engagement patterns. The dual-head ensemble combines these components to jointly predict misinformation engagement levels and the number of new posts per platform-topic pair.
• Figure 3: Platform-specific exchange rates learned by Dreams. Heatmap showing relative likelihood values ($\gamma$ parameters) extracted from FiLM layers for each platform-engagement level combination.
• Figure 4: Fast versus slow memory activation weights from Dreams-Mamba for test instances with MAPE $<10$%. Each point is a correctly predicted instance, colored by different engagement levels, with large markers indicating level-wise centroids. The memory weights are extracted from the dual-timescale memory banks after the forward pass through Dreams-Mamba.
• Figure 5: Emotional dynamics across topics, platforms, and Dreams predictions. (a) Emotional composition of three climate change narratives (Pro-Climate, Climate-Skeptical, and Climate-Anxious) on Facebook (2022), across 28 GoEmotions categories (intensity $0.1$--$0.7$). (b) Emotional flows from aggregated GoEmotions categories through three climate change narratives to predicted engagement levels on Facebook. (c) Emotional flows from aggregated GoEmotions categories through three platforms (X, Facebook, TikTok) to engagement levels for the dog adoption topic (2022).