We use natural language processing (NLP) to detect sentiment. We apply sentiment at two levels: at the document (review) level and the entity level.

We first identify the sentiment of the entire review. Then, once we have identified each entity within the review for the Entity Progress and Top Words components, we calculate sentiment of each separate entity. For example, a review could have positive sentiment overall, but there could be negative entities within the review. The example review below has a document-level sentiment of positive, and the negative entity within is highlighted in red:

The Brandwatch Reviews sentiment algorithm is completely independent from Brandwatch Consumer Research. It's an analysis pipeline built specifically for analyzing reviews, using state-of-the-art multilingual natural language models fine-tuned for machine translation, named-entity recognition, and aspect- and document-level sentiment.