I have noted an odd issue with the way autotags appear to affect search results.
Many autotags have a hierarchal structure. For example to get the TULIP autotag you first have to get FLOWER and to get FLOWER you have to first get PLANT. Similarly, to get the POODLE autotag you first have to get the DOG autotag and prior to that you need ANIMAL.
When you do a text search on ANIMAL or DOG the relevant results typically offer images with high social activity. However, when searching for POODLE, the relevant results offer mainly older images with low social activity. This pattern is noted with PLANT, FLOWER and TULIP as well, as it is with many of these autotag groupings.
Why would high social activity make animal and dogs relevant but low social activity make poodles relevant? Having different algorithms makes perfect sense but not within one hierarchal structure.
So what? For those who search optimise for relevant, it makes no sense to optimise for 3rd level autotags like tulip and poodle, as the standard optimise techniques do not work for the 3rd order tags. Focus on the more popular 2rd order tags.