Document subjectivity classification using a small Italian newspapers articles dataset called SubjectivITA. The dataset is composed of 74 training documents containing 1614 sentences in total, and 29 testing documents, containing 227 sentences.

In order to deal with small data limitation, sBERT, a library containing pretrained, multi-language sentence embedding models can be employed.

Using these pretrained models, one can easily produce meaningful representations for individual sentences, which in their turn can be used as features for simpler classical machine learning models to predict the document class.

This repository includes notebooks for three experiments described in the report file.