I’ll ask the interviewer to clarify what they mean by improving “image search”. Do they mean “google image search” or “image search” in general? I’m going to assume that the interviewer means “google image search”.
I will then ask if interviewer is ok with me using “time it takes to find the right image” as a KPI for improving image search. The shorter this time is, the faster the user finds their image, indicating that the Google Image Search did a good job of finding the image quickly.
Now, I will first chose a particular use case / user group to focus on. Here are a few use cases I can think of:
– looking for picture of a person to see how a particualr (a celebrity, a person from another department at work, a salesperson, etc) looks
– exploring various design options (e.g. dinner table, coffee table, kitchen cabinets, white office chair) for a furniture purchase decision
– people wanting to see how an art work looks like
– looking for the look of something they are curious about (e.g. an animal, a plant, a car model, etc).
For this interview, I will focus on the second use case (exploring design options for a furniture option). I think the current image search experience is limiting for the user because the user cannot limit the search results to cases with a particular home decor in mind. A solution I have in mind is to allow user to upload picture of the environment for which they are buying a piece of furniture and ask google to search for images that have similar decor with a different design of the particular furniture. For example, if google image detects that I have a dark brown hardwood floor, leather couch, black coffee table and i say show me “dinner tables” that go well with this decor, it only shows me images of home decors that have dark brown hardwood floor, leather couch, black coffee table. This way, the user is only presented with meaningful search results, which should reduce the number of searches for the particular furniture design, and time to search.
To implement this solution, google can use machine learning to detect objects, colours, and potentially types of objects to understand the environment around the seeking furniture. Google can then search through images on the internet and only list those that have similar objects, colours, and types of objects.
Can you give me feedback?