This study aims to develop a model for identifying opinion leaders in public safety incidents, with a focus on online public discourse surrounding incidents of telecommunication fraud in Northern Myanmar. The research not only uncovers public attitudes and perspectives on telecommunication fraud, but also explores how to enhance government information management and response strategies through data analysis.
Utilizing Python, we collect and analyze a vast array of user comments and data on Sina Weibo related to telecommunication fraud. By employing an LDA topic model, we categorize and track the evolution of user discussions over time. Furthermore, we calculate text and sentiment similarity to identify opinion leaders within each topic using the PageRank algorithm, creating dynamic graphs that illustrate the temporal evolution of these leaders. The study also delves into the influential characteristics of opinion leaders, establishing a ranking of feature importance, thereby providing a scientific basis for government agencies to identify key influencers when formulating public safety strategies. Most significantly, we develop a machine learning fusion model that combines Gradient Boosting Decision Trees (GBDT) with Multilayer Perceptron (MLP), leveraging deep learning and big data technologies.
This model not only improves the accuracy of identifying opinion leaders, but also bolsters the decision-making capabilities of governments in complex information environments. Through this research, government entities can more effectively monitor and respond to public safety incidents, promptly identify and communicate with opinion leaders, and thus adopt more precise and effective measures in crisis management and public relations.
This study’s limitations include the focus on a single news article’s data and reliance on sentiment dictionaries, which may struggle with evolving online language. Future research should explore diverse data types (e.g. images, videos) and expand sentiment dictionaries to improve accuracy. Additionally, the study’s methods should be tested across various social media platforms to address platform-specific characteristics and enhance generalizability. Future work will also integrate semantic analysis tools like Word2Vec or BERT to overcome TF-IDF limitations and improve text similarity calculations.
The study proposes an enhanced PageRank algorithm incorporating text and sentiment similarity scores to identify opinion leaders dynamically. This approach improves accuracy and robustness compared to traditional methods, offering a valuable tool for large-scale and precise identification of opinion leaders. By integrating content quality and sentiment into network analysis, the model provides actionable insights for public administrators to proactively manage public opinion, especially during public safety events, ensuring effective governance and oversight.
This research highlights the importance of managing public opinion in public safety issues, such as telecommunication fraud, where misinformation can spread rapidly. By dynamically identifying opinion leaders and monitoring public sentiment, the study supports proactive governance and reduces the risk of social panic. The findings promote social consensus and stability by enabling authorities to guide opinion leaders and steer public discourse constructively, especially during emergencies.
(1) Cross-Cultural Sentiment Analysis: The study delves into Sina Weibo data to uncover cross-cultural public responses to incidents of telecommunication fraud. (2) Topic Evolution Analysis: Utilizing an LDA model, the research tracks the dynamic evolution of discourse surrounding telecommunication fraud. (3) Dual-Dimensional Similarity Calculation: Innovatively identifies opinion leaders by combining text and sentiment similarity metrics. (4) Integration of Deep and Machine Learning: Introduces a GBDT-MLP model that synergizes deep learning with machine learning to enhance predictive capabilities. (5) Government Decision-Making Aid: Equips government entities with data-driven support for identifying opinion leaders in public safety incidents.
