Topic Detection and Tracking in Social Media Platforms

The large amount of information available on the Web can be effectively exploited in several domains, ranging from opinion mining to the analysis of human dynamics and behaviors. Specifically, it can be leveraged to keep up with the latest news around the world, although traditional keyword-based techniques make it difficult to understand what has been happening over an extended period of time. In fact, they do not provide any organization of the extracted information, which hinders the general understanding of a topic of interest. This issue can be overcome by leveraging a Topic Detection and Tracking (TDT) system, which allows detecting a set of topics of interest, following their evolution through time. This work proposes a TDT methodology, namely length-weighted topic chain, assessing its effectiveness over two real-world case studies, related to the 2016 United States presidential election and the Covid19 pandemic. Experimental results show the quality and meaningfulness of the identified chains, confirming the ability of our methodology to represent well the main topics underlying social media conversation as well as the relationships among them and their evolution through time.