DT3.8.5 on Monterey using Bonjour sync across 3 desktops and two iOS devices. I have always tolerated some unwanted tags that appear after importing new items into DT3. I recently imported about 50k emails into their own DT3 database. Most acquired tags using the email address as a tag and other unassigned tags that may have been auto-generated by DT3 AI processes. (I noticed the issue because the notification began appearing that I have a huge number of tags potentially slowing Spotlight search.) In my research, I tag some items but rely mostly on DT3 auto-classification to find what I want to find keywords. I would like to keep my tagging to a minimum so that my research databases stay trim without bloat. (Along with unwanted email addresses as tags, some strings of whitespace and alphanumeric characters also appear as tags as do some keywords from within documents.) Is there a pref or hidden feature that can be turned off so that auto-tagging does not occur (and does not bloat the database)?
Auto-tagging is not standard behaviour, so we need to turn off whatever you have (inadvertently?) turned on, I think 
What are your settings in Preferences > Files > Import > Tags?
Selected: Convert keywords to tags, add tags to images, scan barcodes
Deselected: Convert hashtags, properties, geolocations to tags
Follow up question(s):
- Does barcode data become tags?
- If I deselect all, will there be any tags (i.e., are tags and metadata synonymous and/or overlapping)?
- Do I understand correctly that there is already a plethora of metadata and that I can create custom metadata without needing tags?
Yes. You’ll find more information on that (and probably your other questions, too), in the very good manual that is installed with DT.
Tags and metadata are not synonymous. Tags are a type of metadata.
Custom metadata isn’t necessarily a replacement for tags, though this depends on the individual use. Also remember custom metadata is only available in the Pro and Server editions.
Thanks. For my workflow, minimizing auto-created tags and relying on custom metadata for academic research with Server is ideal.
See this blog post of ours…