ABSTRACT The proliferation of the Internet and social networks has significantly expanded the user base of media platforms, making the management of public opinion on these platforms a critical concern. Previous approaches have predominantly examined topics and sentiments either independently or jointly, often limiting the analysis to textual features alone. To address this limitation, this paper proposes a multidimensional “topic‐sentiment‐network” (TSN) method and develops a three‐layer architecture for public opinion analysis. The method integrates BERTopic and SnowNLP for joint topic‐sentiment analysis and constructs complex networks to reveal the structural characteristics of opinion propagation. Finally, a simulation experiment is conducted on the microblog topic “DeepSeek suffering from US‐based brute‐force cyberattacks.” Comparative evaluations demonstrate that the proposed framework achieves superior performance, with a Topic Diversity score of 0.975 and a sentiment analysis accuracy of 71.6%. Furthermore, an ablation study confirms that the sentiment‐weighted network structure improves modularity by 4.2% (from 0.2600 to 0.2708) compared to traditional topology‐based networks. These results indicate that this method provides a systematic, comprehensive, and quantitative framework for monitoring public opinion dynamics.
Zexi et al. (Wed,) studied this question.