Learning Algorithms of Sentiment Analysis: A Comparative Approach to Improve Data Goodness

The uncritical application of automatic analysis techniques can be insidious. For this reason, the sci entific community is very interested in the supervised approach. Can this be enough? This chapter aims to these issues by comparing three machine learning approaches to measuring the sentiment. The case study is the analysis of the sentiment expressed by the Italians on Twitter during the first post-lockdown day. To start the supervised model, it has been necessary to build a stratified sample of tweets by daily and classifying them manually. The model to be test provides for further analysis at the end of the pro cess useful for comparing the three models: index will be built on the tweets processed with the aim of detecting the goodness of the results produced. The comparison of the three algorithms helps the authors to understand not only which is the best approach for the Italian language but tries to understand which strategy is to verify the quality of the data obtaine