depth-classifier-updater

To regenerate keyword lists for a depth classifier from MongoDB conversations, run the update script to sample recent user messages, classify them as SHALLOW or DEEP, and deduplicate phrases to refresh the classifier file.

You should update your depth classifier during routine maintenance, after feature deployments, or whenever classifier accuracy drifts. This realigns SHALLOW and DEEP routing decisions with current real user conversation patterns.

Yes, you need a MongoDB connection, the Kazusa AI Chatbot package, and a Python runtime to execute the update script, re-sample messages, balance languages, and run tests for the depth classifier.

The depth classifier updater balances bilingual keyword coverage by classifying sampled messages using signal-based heuristics, then deduplicating phrases to ensure balanced English and Chinese keywords in the updated classifier file.

The best way to validate depth classifier keyword updates before deployment is to run the integrated tests within the Python script after regenerating the SHALLOW and DEEP lists, which provides deployment guidance and confirms accuracy.

You need to sample 300 to 1000 recent messages from MongoDB to update the depth classifier. This volume ensures the regenerated keyword lists accurately reflect current usage and improve routing decisions.