What to do about the algorithm being influenced by an excessive quantity of incorrect observations of a species?

Unfortunately, there’s no current solution aside from manually re-identifying them. However, doing this does seem to improve the suggestions pretty quickly.

In cases that don’t require any specialist knowledge, like the ‘wrong continent’ example you give, it’s often simplest to write a short journal post explaining your reasoning, and invite other identifiers to help with corrections. People can just post a link to your explanation, and it will go pretty quickly.