103,482 scraped Swedish-language reviews balanced across positive and negative sentiment classes. The collection is partitioned into training, validation, and test sets derived from a raw source originally skewed 95/5 toward positive samples.
Use Cases
- Train a binary sentiment classifier using the review text and sentiment labels
- Benchmark Swedish language models on sentiment detection using the provided test split
- Analyze linguistic patterns in negative Swedish consumer feedback compared to positive reviews
Strengths
- 103,482 samples across train, valid, and test splits
- Balanced class distribution targeting the 5% minority negative class from the original source
- Aggregated from multiple Swedish websites to capture diverse consumer review styles