• cluster analysis
• coronavirus
• covid-19
• stigmergy
• topic detection
• topic extraction

Nowadays, more than half of the world’s population is an active social media user: billions of people of every age, every social class and from all over the world use social media platforms to inform themselves and for sharing their own contents. This has led to the creation of large amount of data, produced every day especially through microblogs, online mediums that allow users to easily share contents of small size such as short sentences, photos or videos. Huge quantities of information are produced in particular during exceptional events like natural disasters, terrorist attacks or sport matches.
This thesis proposes an approach that uses Twitter messages for detecting topics related to an event by identifying all the terms related to it and extracting meaningful information from them. The most used words are processed using a stigmergic approach that evaluates the similarity of their dynamics in time with respect to some archetypal behaviors. Then, clustering algorithms are employed in order to summarize the different episodes occurred during the event, according to the similarities of their terms.
In this work, data was obtained through Twitter API and tweets about the COVID-19 pandemic were collected between August 16th, 2020 and August 17th, 2020. The system was proven to be able to detect all the main topics of discussion emerged in Italy during different moments of one of the most eventful day of the pandemic.