Concept-Based Visual Analysis of Dynamic Textual Data (2306.10462v1)

Abstract: Analyzing how interrelated ideas flow within and between multiple social groups helps understand the propagation of information, ideas, and thoughts on social media. The existing dynamic text analysis work on idea flow analysis is mostly based on the topic model. Therefore, when analyzing the reasons behind the flow of ideas, people have to check the textual data of the ideas, which is annoying because of the huge amount and complex structures of these texts. To solve this problem, we propose a concept-based dynamic visual text analytics method, which illustrates how the content of the ideas change and helps users analyze the root cause of the idea flow. We use concepts to summarize the content of the ideas and show the flow of concepts with the flow lines. To ensure the stability of the flow lines, a constrained t-SNE projection algorithm is used to display the change of concepts over time and the correlation between them. In order to better convey the anomalous change of the concepts, we propose a method to detect the time periods with anomalous change of concepts based on anomaly detection and highlight them. A qualitative evaluation and a case study on real-world Twitter datasets demonstrate the correctness and effectiveness of our visual analytics method.