The rapid proliferation of digital technologies and the exponential growth of online content have underscored the critical importance of identifying and managing cyber risk public opinion efficiently. This study, grounded in the burgeoning field of artificial intelligence (AI), presents a novel approach aimed at the identification and emergency governance of cyber risks and public opinions. It harnesses the power of Long Short-Term Memory (LSTM) networks enhanced by an innovative optimization strategy known as the Tuna Algorithm (TA), marking a significant leap forward in the precision and efficiency of online public sentiment analysis. The research delves into the intricacies of online public opinion, highlighting its dual nature: as a reflection of public sentiment and a potent force capable of rapidly influencing public discourse and decision-making. In an era characterized by information overload, the agility with which online public opinion forms and spreads poses both opportunities and challenges. The digital public sphere, encapsulating diverse viewpoints and emotional expressions, becomes a double-edged sword—offering platforms for engagement and dialogue but also paving the way for misinformation and the escalation of cyber risks. Based on the nuanced complexities of online sentiment, this study employs LSTM, a form of recurrent neural network adept at handling sequence data over long periods. The choice of LSTM is strategic, aiming to capitalize on its proficiency in capturing temporal dependencies and contextual nuances inherent in the flow of online public opinion. However, the conventional LSTM model, despite its strengths, encounters limitations in addressing the dynamic and multifaceted nature of cyber risk public opinion. To transcend these boundaries, the research integrates the Tuna Algorithm (TA), a heuristic optimization method inspired by the foraging behavior of tuna. The TA’s global search capabilities and rapid convergence enhance the LSTM model, enabling it to swiftly adapt to the evolving landscape of online sentiment. The hybrid LSTM-TA model proposed in this study demonstrates a remarkable accuracy rate of approximately 97% in processing and analyzing data, underscoring the model’s enhanced capability to identify and manage online risky public opinions accurately. This leap in performance is not merely a technological advancement but a beacon for future research and applications in cyber risk public opinion management. The model’s success lies in its meticulous construction, where the LSTM’s analytical depth is amplified by the TA’s optimization prowess, showcasing a synergistic blend that optimizes performance metrics significantly. Experimentation and evaluation form the bedrock of this research, with diverse datasets spanning hotel reviews, social media comments, and consumer feedback on e-commerce platforms serving as the empirical basis. These datasets, reflective of varied domains and scenarios, provide a comprehensive testing ground for the LSTM-TA model. The study’s methodical approach to model evaluation, comparing its performance against traditional and contemporary models in AI, underscores the LSTM-TA model’s superior accuracy and efficiency. Such comparative analysis not only validates the model’s effectiveness but also illuminates its potential to revolutionize the identification and governance of cyber risk public opinion.