Scholarlink

<p>Public opinion crises can rapidly escalate in the digital age, affecting the reputation and stability of organizations and governments. Timely and intelligent response strategies are crucial for effectively containing and neutralizing such crises. However, existing approaches often rely on static communication templates and delayed manual decisions, which lack adaptability to evolving sentiment and the rapid dissemination of misinformation. To address these challenges, this study proposes a Deep Reinforcement Learning-based Crisis Response Optimization (DRL-CRO) framework that dynamically adjusts communication strategies in real-time. The framework models the crisis environment using state variables, including public sentiment, information dissemination, and platform activity. The action space encompasses content type, tone, platform selection, and response timing. By optimizing a reward function based on trust restoration and misinformation reduction, the DRL agent continuously learns and improves its response strategies. The method was evaluated in a simulated public opinion crisis scenario and compared against baseline models, including MLA, CNN, and FLA approaches, to provide context for performance assessment. This method is applied to a simulated public opinion crisis scenario, where real-time social media data guides adaptive decision-making. Experimental results show that DRL-CRO significantly outperforms traditional rule-based systems in terms of sentiment recovery, misinformation containment, and communication engagement. The findings demonstrate the potential of reinforcement learning in enhancing public crisis management with adaptive and intelligent response mechanisms. The proposed method gradually improves the sentiment recovery rate by 98%, misinformation containment by 97.3%, response latency by 63%, engagement rate by 756%, trust restoration index by 97.4%, platform strategy effectiveness by 94.1%, reward convergence stability by 3.2%, and comparative performance index by 92%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic response and disposal strategies for public opinion crises driven by reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月3日

查看全文
  • Yang Chu

摘要

Public opinion crises can rapidly escalate in the digital age, affecting the reputation and stability of organizations and governments. Timely and intelligent response strategies are crucial for effectively containing and neutralizing such crises. However, existing approaches often rely on static communication templates and delayed manual decisions, which lack adaptability to evolving sentiment and the rapid dissemination of misinformation. To address these challenges, this study proposes a Deep Reinforcement Learning-based Crisis Response Optimization (DRL-CRO) framework that dynamically adjusts communication strategies in real-time. The framework models the crisis environment using state variables, including public sentiment, information dissemination, and platform activity. The action space encompasses content type, tone, platform selection, and response timing. By optimizing a reward function based on trust restoration and misinformation reduction, the DRL agent continuously learns and improves its response strategies. The method was evaluated in a simulated public opinion crisis scenario and compared against baseline models, including MLA, CNN, and FLA approaches, to provide context for performance assessment. This method is applied to a simulated public opinion crisis scenario, where real-time social media data guides adaptive decision-making. Experimental results show that DRL-CRO significantly outperforms traditional rule-based systems in terms of sentiment recovery, misinformation containment, and communication engagement. The findings demonstrate the potential of reinforcement learning in enhancing public crisis management with adaptive and intelligent response mechanisms. The proposed method gradually improves the sentiment recovery rate by 98%, misinformation containment by 97.3%, response latency by 63%, engagement rate by 756%, trust restoration index by 97.4%, platform strategy effectiveness by 94.1%, reward convergence stability by 3.2%, and comparative performance index by 92%.