See how Wikibots are working at Wikipedia, it is a good reference model… Other important criteria to policy and implementation:
-
to be an assisted agent: never (or very rare) an automatic classification with no human confirming it. Must work “by (assisted) lot” not in a continous (by demand) way.
-
to be specialized: never generic, need a scope and limited (configurable/controlable) range.
-
to be symmetric over validation: an human can be also use the same tag, the robot (software agent) is not the only autorithy to tagging its scope. So two cenarios: 1. robot tagging and human validating/confirming (or warning to vote). 2. human tagging, and robot valiating (or warning to other human confirm).
Whith this basic polices the robot is viable in any community!
Other suggestion: when the scope is specific, or when the “easy to tag” generates “tag pollution”, the best is to use different color to the tags of the bot-scope. EXAMPLE: tagging languages in a multilingual community as https://discuss.okfn.org.