That said, "is it hacking or not" isn't really a terribly interesting meta-discussion to have, imho, so I'll try and go in a different direction.
Please, if that isn't flamebait I really don't know what is.
The data appears to support the assertions you are calling flamebait. Provided their algorithm for calculating likeability actually works, then the data does support the assertions[1]. (This is all, of course, within the context of the pre-selection you noted in your post).
[1] Note that the assertions are of the form "Members of Set X are more likeable to members of Set Y than members of Set Z". If the assertions were "Being a member of Set X makes you more likeable to members of Set Y", they wouldn't necessarily be supported by the data.