Filmweb+ is a dataset of Polish movie reviews used for the purpouse of a document-level sentiment classification task.
It contains 27202 reviews in total, which are divided into unlabelled, positive, neutral and negative classes based on rating given by the review's author.
| Class (indice) | # of reviews |
|---|---|
| Unlabelled (-1) | 7710 |
| Negative (0) | 3085 |
| Neutral (1) | 9436 |
| Positive (2) | 6971 |
Reviews have been fetched from Filmweb, Film.org.pl, Kinoblog, FDB and blogfilmowy24. They contain an average of 514 words (or 3582 characters) per document.
The dataset has been pre-processed in order to remove unnecessary HTML tags or other redundant metadata, such as image descriptions, using the Spacey preprocessor. It is available in a CSV format.
A ULMFiT-based document level sentiment classification system proposed in work "Sentiment Classification using Machine Learning" (Narolski, 2020) has achieved a state-of-the-art result of 94,19% for Polish document-level sentiment classification task for a corpora of long, real-world text documents, with over 500 words per document on average.
If you've found this dataset helpful for your work, please consider citing the work "Sentiment Classification using Machine Learning" (Narolski, 2020) or providing an appropriate authorship attribution.