Misinformation Regulation in the Presence of Competition between Social Media Platforms (Extended Version)

TL;DR

Effectiveness of regulation depends on the diffusive property of news posts, friend interaction qualities in social media, the sizes and cohesiveness of communities, and how much sympathizers appreciate surprising news from influencers.

Abstract

Social media platforms have diverse content moderation policies, with many prominent actors hesitant to impose strict regulations. A key reason for this reluctance could be the competitive advantage that comes with lax regulation. A popular platform that starts enforcing content moderation rules may fear that it could lose users to less-regulated alternative platforms. Moreover, if users continue harmful activities on other platforms, regulation ends up being futile. This article examines the competitive aspect of content moderation by considering the motivations of all involved players (platformer, news source, and social media users), identifying the regulation policies sustained in equilibrium, and evaluating the information quality available on each platform. Applied to simple yet relevant social networks such as stochastic block models, our model reveals the conditions for a popular platform to enforce strict regulation without losing users. Effectiveness of regulation depends on the diffusive property of news posts, friend interaction qualities in social media, the sizes and cohesiveness of communities, and how much sympathizers appreciate surprising news from influencers.

Figures (8)

• Figure 1: Illustration of our model. Blue (red) nodes are in platform A (B). Values below nodes present $p_{i\mathbf{P}_i}$. Note that user 6 is assumed to receive the source's signal with probability $p^2$ (because it is two edges away from it) even though there are two paths connecting these vertices. See text for details.
• Figure 2: The sender's expected utility in a linear network
• Figure 3: The strictest effective regulation $\rho_{SE}$ in (a) linear network, (b) star-chain network, and (c) regular-tree network. Respectively, (d), (e), (f) are the finite cases. In panel (a), the white curve is computed using the formula of Proposition \ref{['bA_threshold_for_strictest_restriction']}. From this proposition, the strictest effective regulation is guaranteed to be $\rho_{SE}=0$ above this curve in the $(p,b_\mathbf{A})$-plane. The orange curve plays the same role, but based on the results of Proposition \ref{['prop_strict_regulation_linear']}. The color map is computed by \ref{['Ustar_B']} and Proposition \ref{['strictest_regulation']}. The white/orange curves and color maps in other panels are computed similarly.
• Figure 4: The strictest effective regulation $\rho_{SE}$ for (a1, a2) community chains and (b1, b2) complete graphs of communities. The community chain has a middle community of (a1) low or (a2) high intra-community tightness. Parameter values are as described in the main text. The color map is the simulation results of 50 random graphs. The white curves calculated by Proposition \ref{['thm_sbm_thresh']} fit the results of the simulation, which does not use Assumptions \ref{['assumption_community_migration']}--\ref{['assumption_external_friend']} for the proposition. The dashed curves represent the base case with medium intra-community tightness (see Figure \ref{['fig:regulation_differentc_communitychain']} panel (b)). With high $p$, platform $\mathbf{A}$ can enforce strict regulation more easily (i.e., the white curve is lower) in (a2) than (a1). For the complete graphs of communities, the sender is in a (b1) small or (b2) large community. The white curves calculated by Proposition \ref{['thm_sbm_thresh']} again fit the color maps (the simulation results for 50 random graphs), which are computed without Assumptions \ref{['assumption_community_migration']}--\ref{['assumption_external_friend']} for the proposition. With low $p$, platform $\mathbf{A}$ can enforce strict regulation more easily in (b2) than (b1).
• Figure 5: The strictest effective regulation $\rho_{SE}$ for infinite linear networks. Panel (b) is the basic case with $c_i=0.3$ for all $i$. For panel (a), users close to the sender are sympathizers ($c_i=0.21$ for $i=0,1,2$) and distant users are non-sympathizers ($c_i=0.4$ for $i\ge 3$). Panel (c) is the opposite ($c_i=0.4$ for $i=0,1,2$ and $c_i=0.21$ for $i\ge 3$). The white curves are the parameter threshold for $\rho_{SE}=0$ as in Proposition \ref{['bA_threshold_for_strictest_restriction']}, and the dashed curves represent that in the base case (b) for reference. In case (a), it is difficult for platform $\mathbf{A}$ to enforce strict regulation (i.e., the color map is bright and the white curve is high) when $p$ is low. The color map is computed similarly to Figure \ref{['fig:regulation']}, not using any random procedure.

Theorems & Definitions (19)

• Proposition 1
• proof
• Definition 1
• Proposition 2
• Definition 2
• Proposition 3
• Proposition 4
• Proposition 5
• Proposition 6
• Proposition 7