Sentiment Classification is one of many AI driven models running on Sonar. The model automatically attaches a 👍 Positive, 👎 Negative or ✋ Neutral tag on all quotes. It's a complex classifier and it's running on a state of the art AI model but the output is simple, as it classifies if your customer is positive, negative or neutral when they talk about your ideas or products. As with any other AI generated output on Sonar, you can also change the sentiment, which will help the classifier to learn from its mistakes and improve its accuracy in the future.

How it works

Sentiment classification works by predicting if a section of text is positive, negative or neutral. The model not only looks at single words, but each word in context of the rest of the text section which makes it very reliable for sentiment analysis. For each word the classifier will predict sentiment with a certain level of confidence. If the level of confidence across words in the text section reaches a threshold, the model will attach a corresponding tag to the quote nugget.

Fine-tuning state of the art AI technology

The sentiment classifier is a proprietary model based on BERT (Bidirectional Encoder Representations from Transformers) - a language model originally developed by Google. Sonar's AI team has trained the model on thousands of open source review examples and years of historical domain specific data, to make sure it understands exactly how to predict sentiment on CX data.

Supported languages