A sentiment reporting framework for major city events: Case study on the China-United States trade war
2020 Elsevier Ltd Smart cities are conceptualized as a vehicle for sustainable urban development and a means to deliver high quality of life for residents. One of the core functions of a smart city consists in the continuous monitoring of events, assets and people and the use of this information and intelligence for the streamlining of the city's operations. Public opinion represents one type of intelligence of particular importance and value. By monitoring public opinion, governments seek to understand prevalent views about the current events and policies, as well as identify extreme views and trends that may represent problematic situations or precursors to violent actions. Ultimately, maintaining a constant awareness of public opinion means that authorities can better assess and predict public reactions in relation to ongoing events, and thus take appropriate actions to maintain public safety. Due to the popular use of social media to express sentiments and emotions about current events, social media content analysis has been contemplated as a promising solution to capture public opinion. However, existing approaches take a coarse-grained retrospective approach to social media content analysis. Furthermore, those approaches suffer from the lack of scalability and efficiency, as they necessitate the collection and analysis of large volumes of social media content (often millions of posts), to come up with relevant conclusions. In this work, we address those limitations by proposing a novel framework for the real-time monitoring of public opinion. To ensure efficiency and scalability, our framework focuses on the analysis of high impact social media content generated by opinion leaders and their followers as means to offer in-depth insights and sentiment intelligence reports about events, as they are occurring in real time. The proposed framework was implemented and tested on data harvested from 52 economic opinion leaders, with a focus on the China-US trade war as case study. The results show that the convolutional neural network (CNN) classifier used for sentiment analysis yielded a classification accuracy of 86% when differentiating between four sentiment categories: Support, strong support, dissent, and strong dissent. The Support Vector Machine (SVM) classifier employed to perform in-depth emotional analysis attained an accuracy of 82% when differentiating between five emotions: Angry, depressed, excited, happy, and worried. Unlike existing retrospective social media analysis approaches that require the analysis of millions of posts, our approach focuses on the analysis of high-impact social media content in real-time, thus constituting an efficient, sustainable, and timely solution to public opinion monitoring.