Investigating the Sentiment in Italian Long-COVID Narrations

Through an overview of the history of the disease, Narrative Medicine (NM) aims to define and implement an effective, appropriate and shared treatment path. In the context of COVID-19, several blogs were produced, among those the “Sindrome Post COVID-19” contains narratives related to the COVID-19 pandemic. In the present study, different analysis techniques were applied to a dataset extracted from such “Sindrome Post COVID-19” blog. The first step of the analysis was to test the VADER polarity extraction tool. Then the analysis was extended through the application of Topic Modeling, using Latent Dirichlet Allocation (LDA). The results were compared to verify the correlations between the polarity score obtained through VADER and the extracted topics through LDA. The results showed a predominantly negative polarity consistent with the mostly negative topics represented by words on post virus symptoms. The results obtained derive from three different approaches applied to the COVID narrative dataset. The first part of the analysis corresponds to polarity extraction using the VADER software, where, from the score, polarity was inferred by dichotomizing the overall score. In the second part, topic modeling through LDA was applied, extracting a number of topics equal to three. The third phase is based on the objective of finding a qualitative relationship between the polarity extracted with VADER and the latent topics with LDA, considering it a semi-supervised problem. In the end, the presence of polarized topics was explored and thus a correspondence between sentiment and topic was found.