SentiLexIT: Advancing Italian Sentiment Analysis through Automated Lexicon Generation

Abstract

The proliferation of digital platforms has exponentially increased the volume of user-generated content, providing rich data sources for sentiment analysis. This study introduces SentiLexIT, a novel methodology for automatically constructing domain-specific sentiment lexicons for the Italian language using an unsupervised approach. We analyzed 321,817 user reviews from diverse categories such as Amazon.it and Google Reviews, applying Bayes' Theorem to derive sentiment scores that effectively categorize review sentiments into distinct polarities. Our methodology significantly outperforms existing Italian sentiment lexicons and two versions of ChatGPT, achieving an average F1-score of 0.92 across various domains. The paper highlights the method's effectiveness in handling linguistic nuances specific to Italian, providing a robust tool for sentiment analysis. This research contributes to the field by offering a scalable, precise methodology for sentiment lexicon construction without the need for manual annotation, addressing the gap in non-English sentiment analysis resources.