About
Traditionally, sentiment scores have been limited by simple, sentence-based models that don't truly capture the dynamics of a conversation. These models only reflect the type of words used, like "good", "bad" or "sorry," without considering the overall context.
Dubber's new Conversation Sentiment score uses state-of-the-art language models to analyse entire conversations. It considers speaking turns and full context, providing a more accurate picture of how sentiment evolves throughout an interaction.
We've thoroughly tested this new approach using internal datasets and subjective user evaluations. Results confirm that our new scores more accurately reflect speaker sentiment in real conversations.
More Information
For more information see: Conversation Sentiment Uplift