H-NeiFi: Non-Invasive and Consensus-Efficient Multi-Agent Opinion Guidance

TL;DR

The paper tackles the problem of steering online opinion toward global consensus without compromising user autonomy, addressing echo chambers and the downsides of invasive interventions. It introduces H-NeiFi, a hierarchical, non-intrusive framework that combines PCP (for expert-to-expert alignment) and ACP (a MARL-driven, long-term neighbor filtering mechanism for non-experts) with a dual reward design to balance local cohesion and global convergence. The method leverages a bounded-confidence opinion dynamics model, a two-layer dynamic structure, and a long-horizon MARL objective to optimize information propagation paths rather than altering user opinions directly. Across synthetic experiments with varying scales and expert presence, H-NeiFi achieves faster and more robust global consensus (22.0%–30.7% acceleration in certain settings) while maintaining convergence, highlighting its potential for natural, autonomy-preserving social network governance.

Abstract

The openness of social media enables the free exchange of opinions, but it also presents challenges in guiding opinion evolution towards global consensus. Existing methods often directly modify user views or enforce cross-group connections. These intrusive interventions undermine user autonomy, provoke psychological resistance, and reduce the efficiency of global consensus. Additionally, due to the lack of a long-term perspective, promoting local consensus often exacerbates divisions at the macro level. To address these issues, we propose the hierarchical, non-intrusive opinion guidance framework, H-NeiFi. It first establishes a two-layer dynamic model based on social roles, considering the behavioral characteristics of both experts and non-experts. Additionally, we introduce a non-intrusive neighbor filtering method that adaptively controls user communication channels. Using multi-agent reinforcement learning (MARL), we optimize information propagation paths through a long-term reward function, avoiding direct interference with user interactions. Experiments show that H-NeiFi increases consensus speed by 22.0% to 30.7% and maintains global convergence even in the absence of experts. This approach enables natural and efficient consensus guidance by protecting user interaction autonomy, offering a new paradigm for social network governance.

Figures (5)

• Figure 1: Illustration of opinion evolution in a social network. (a) illustrates the initial opinion distribution, (b) demonstrates isolated clusters formed by the echo chamber effect. During the evolution process, the convergence of existing opinion baselines and our method is shown in (c) and (d). The baselines rely on direct influences, resulting in localized consensus. In contrast, our H-NeiFi achieves global consensus through hierarchical communication patterns and a non-intrusive neighbor filtering method.
• Figure 2: H-NeiFi framework. At the start of each training round, a global goal is set and users are categorized into non-experts and experts. PCP filters neighbors for experts based on the minimum disagreement, while ACP filters neighbors for non-experts based on MARL for long-term planning.
• Figure 3: Illustration of opinion evolution processes without expert influence ($x_{v_i}(0) \in [0,8]$ and $m = 80$).
• Figure 4: Opinion evolution processes under uneven opinion distributions.
• Figure 5: Opinion guidance processes in multi-expert settings.