During the Discover phase, why is it recommended to combine cluster labeling with search-based labeling?
Cluster labeling automatically groups semantically similar messages, making it efficient for identifying and labeling high-frequency, common patterns in the data. However, clustering alone can miss rare or less frequent message variations that don't form large enough clusters. Search-based labeling complements this by letting you specifically search for and label these rarer or edge-case utterances. Using both techniques together ensures the training dataset has broader coverage of both common and uncommon message patterns, which results in a more robust and precisely trained model.
Community Discussion