The aim of this research work is to fill a significant gap in the existing literature about the application of NLP for language other than English: to present a comprehensive deep learning model for text categorization in the Italian language. The experiments conducted on a corpus of real-life Italian texts show that the best precision of classification (91.7%, up to 14.3% from the initial models) is obtained by combining ELMo contextualised insertions with Bidirectional – Long Short-Term Memory (BiLSTM) netwks. This study meets a gap in English-specific NLP by providing practical solutions to global problems of text categorisation, in-dpth analysis of architectures and repeatable guidelines.