Sentiment analysis model based on BERT, fine-tuned on product reviews.

Trained on a proprietary dataset of 500k product reviews. Input is a text snippet, output is a sentiment label (positive/neutral/negative).

Designed for analyzing customer feedback in the Aurora retail platform. Not intended for medical, legal, or financial decisions.

May exhibit bias toward informal language. Performance degrades on non-English text. Users should validate on their own data.

Trained on Aurora internal review data collected between 2023 and 2025.