Consider associations when making suggestions

Wasabe

Suppose I have 100 pictures of Mr. X, and in 80 of them, Mrs. X is standing beside him. And there are no pictures of Mrs. X by herself. This information could be useful when suggesting possible matches for faces.
I mean, the unrecognized face in a picture that has Mr. X is much more likely to be Mrs. X than an unrecognized face in a picture that doesn’t have Mr. X

Some combinations of people simply don’t occur. My colleague in Tokyo is never in the same picture with my cousin in New York, for example. Considering which people occur together and which don’t can help produce more reasonable suggestions for unrecognized faces.

The same concept could also be applied to dates. If the only pictures of a certain person are from 2018 onward, then it is unlikely that they are the unknown person in a picture from 1995.

Andrey

Wasabe Dates are already used as one of the features during similar faces suggestion. However, company is not. Interesting idea.

Wasabe

Location is another potential factor that can be used this way.
I have previously developed a similar application (non-commercial, for my own use), and I used this approach.
I found it was very effective in making the suggested matches for an unknown face more useful.

From my experience, there is some benefit to associations by date, and some for location, but the largest benefit (by far) was from associations with other people occuring in the same pictures.