I've done quite a few attempts at this, and currently I've kind of narrowed it down to the following in order to find most liked topics that have more relevance than simple first post likes or total of likes in the thread.
num_distinct_likers + (num_distinct_likers / num_messages_liked) + min(num_distinct_likers, (1 / log(total_like_count / (num_replies + 1))))
The quick breakdown is as follows
distinct_likers
Should be obvious, the more unique members that like a thread the more popular it should be.
num_distinct_likers / num_messages_liked
This rewards threads that generate the distinct likers in fewer posts. So if a thread generated 20 unique likes in 3 posts vs 20 in 20 posts it would get more weight.
min(num_distinct_likers, (1 / log(total_like_count / (num_replies + 1))))
This is limited to never add more weight than the the first term above. The 1 / log() section rewards threads that generate a total like count in fewer replies. So again adds weight to threads that have posts with many likes vs threads with many posts with a bunch of single likes. This term can also be negative, so if you have a thread with a lot more replies than likes it will remove weight.
So not perfect by any means but does seem to have the effect of finding the more meaty threads.