Kantar’s Dx Analytics practice has built a novel approach to sentiment classification that has dramatically improved accuracy compared to any other automated sentiment tool with the help of artificial intelligence. The addition of dynamic prompting ensures we can differentiate sentiment responses for each statement and brand you are interested in measuring. This new technology positions Kantar to measure conversations around nuanced topics and sensitive issues with 39% more accuracy than other market-leading social listening platforms.
The power of Kantar’s new generation of AI models allows DX Analytics to:
- Separately judge the sentiment towards each brand/entity being discussed. Our technology reads social statements from each entity’s perspective. This context adds significantly to classification accuracy for statements which compare brands, which are often left as “mixed” or “neutral” in standard automated sentiment tools.
- Customize sentiment judgment to the brand context. This prevents context-specific interpretation errors. For example, the Kantar methodology makes sure words such as “breakup” or “heartbreak” are not coded as negative in conversations about brands like Ben & Jerry’s, when they should be positive.
- Recognize indirect sentiment towards a given brand/entity through context. This allows the model to know that sentiment towards “Cherry Garcia” is a reflection on Ben & Jerry’s without an analyst needing to maintain long flavor/SKU taxonomies.
Dx Analytics builds cutting-edge AI models across Kantar products, thanks to constant exploration and agile implementation of new methodologies. We are periodically testing a variety of LLMs, and these foundational models have improved Kantar's “plug-and-play” capabilities. For example, Kantar’s Q&A interface doesn’t require developers to learn all new syntax to test new models.
Kantar’s Dx Analytics practice measures and optimizes the accuracy of our sentiment models by comparing the automated results with the results from expert human review. Kantar defines the accuracy of a sentiment model with a blend of three key metrics - the classification F1 score, the error in Net Sentiment Scores (NSS), and the degree of Sentiment Inversions.
To learn more about our approach to social listening/data mining, how we calculate social attention and attitudes, and how AI is impacting it all, download our booklet here.