A Machine Thematic Classification Method and Its Application in Automatically Analyzing Social Media Content

With the large volume of comments regarding a societal issue on social media, content analysis surpasses human capacity. Thus, an efficient and reliable machine thematic classification method is greatly needed to provide meaningful information to stakeholders for making proper decisions. In this research, two studies are included. In Study 1, we developed a machine thematic classification method capable of classifying comments into themes based on user-designated classification framework. The method involves processes such as preprocessing comments, establishing a categorization framework, creating a training set by labeling comment themes, model training and machine thematic classifying each comment. The method achieved 74.64% accuracy rate in correctly classified instances and demonstrated competitive model performances. In Study 2, the developed method was tested on 665,298 comments posted on FB pages of public health authorities (PHAs) of Taiwan, Singapore, the United States, and the United Kingdom (S4). The real-world testing has shown that this developed machine thematic classification method can reveal meaningful information for stakeholders that would otherwise be impossible to obtain. This method has the potential to extend to other fields (such as business, science, industry, etc.) to improve or develop strategic analysis, product development, public opinion management, etc.