Each vCA can upload the CSV file of reviews and interactively and in a composing manner can flag reviews. That is his initial work gets probabilistically multiplied over the remaining of his work. That is as they flag reviews early with high attention span they train the spam filter. As their initial flagging constructs a model of what spam and bot reviews look like the filter will just automatically flag them away.
vCAs will have more focus time to work on quality reviews
A 2 category spam filter might not be enough for the complexity of reviews. Yet at least it must be good enough to get spam reviews away. Or do you still receive a lot of spam in your email inbox?