Does it make sense to mess around with neural networks?

Question

Neural networks

There is a task. At the entrance there are many photographs of portraits, which are made in different styles. A person chooses, for example, a dozen that he likes, and the program gives him recommendations from a set of photos.
Of course, you can manually register some tags, but there can be a lot of them, and it may not be known in advance by what criteria a person chooses photos.
Can this problem be solved with the help of neural networks, preferably with the help of more or less ready-made tools. Unfortunately, I know little about them, but it seems to me that this is not a very good idea, since the implementation will require manual training of a neural network.

Answer 1

A person may like a photo because of:
subject matter, artistic style, set of colors (pink penguin), unusualness, relevance and other factors.
This or that photograph can have a psychic effect. I don't think there's any way to measure that without scanning brain activity during the show.
In other words, I have no idea how to train an artificial neural network to select something or, moreover, let it decide for itself.
But within the framework of the study - try it. Tell me later how it turned out.

Answer 2

Yes, this is the mainstream approach. Is it possible to decompose the task, for example, to teach one neural network to tag, and then include it in that one? which makes recommendations.

Answer 3

If you do not have a million users who will somehow group the picture, then it will not take off.