We’re continually making improvements to Answers search term clustering to make it even easier to identify what your customers are searching for on your experience.
In this release we made two major updates to our cluster generation process to help better identify clusters:
- We updated the embedding model, which is the machine learning model that compares the semantic meaning of search terms.
- We are now selecting the epsilon (how strict our clustering algorithm is in terms of grouping search terms together) and the overlarge cluster threshold (threshold where clusters are too large and are discarded) based on the size of the Answers experience (the number of search terms your experience receives), which should improve the accuracy and percentage of search terms that are clustered.
For example, these improvements made to clustering will ensure that search terms like “auto loans,” “car loans,” and “vehicle loans” - while different terms - are clustered together as they all contain the same underlying user intent.
To learn more about search term clusters, visit the Search Term Clustering training unit.
Turn on the Spring ‘22: Clustering Enhancements (early access) account feature to use this feature during the Early Access period. This feature will automatically be turned on for all accounts at General Availability for the Spring '22 Release.