To make a long story short, it seems like the web_detection portion of the vision API is not returning particular entities even though I'm 100% sure that they are supposed to be there and I can't find any documentation of this change or anyone who's talked about it anywhere.
As far as I can tell, the entities in question seem to be solely people, but that could just be a bias based on the type of data I'm working with. It also only seems to be quite important people, but, again, could be my data.
So far, the list of people that refuse to 'show up' are: Donald Trump, Hillary Clinton, Barack Obama, Joe Biden. In contrast, Mike Pence is showing up fine, so is Bill Clinton and Rudy Guilliani.
It gets weirder, though: 'Donald Trump Jr.' shows up just fine, as does the entity 'Donald Trump's 2020 election campaign' - this applies to the others as well. It is only the =='Donald Trump' entity that appears to be missing
I should clarify my method: I'm testing by re-running 120 images of data using web_detection that I (1) know includes the entity (Donald Trump) and (2) ran in 2018 and got a comparable output but with the entity in question as the highest-ranking entity by confidence.
I also ran a small testing batch of about 50 images of Mike Pence and as expected he turns up as the highest-confidence entity in almost all of them. From this, I am fairly confident that the error is not on my part.
Does anyone know why this is happening or has anyone experienced anything like this with their own data? Since it only pertains to a very select group of people, I would assume that this either a strange error related to an internal change or a deliberate decision related to some abuse of the system during the election period.