Topic: What do you think of this idea: An "Aesthetic Searcher" in e621...

Posted under General

What about this...

Imagine a procedure, a program, that search in your favs for the most "significant" tags, and preference rank, and then does a search for a "conjunction" of those tags, and maybe some other characteristics, and retrieves other posts that potentially you may like, nd maybe with some limiting criteria or order (such as date of post).

For "significant" I would mean "statisticaly significant".

----------------------------

The tag system in this site, for what I know that it is very useful to search for an image that you already have seen, with certain "objective" characteristics. I think that with the proper practice, I works just extrimely fine for that.

What about, retrieving posts (images), that you have never seen, with a high probability that you will like too, in a "subjective" sense, aesthetically?

Thinking about it, the solution, at least for me, does not seem to me so evident... is it?

=(.

But intuitively, I am almost sure that something like that could be done.

I will state it like this:

Solve the underlying problem:

"Is there a program, maybe a IA program, that could search, based in mine or other peoples favs, via tags and meta-data, similar pictures that I will probably AESTHETICALLY like?"

=).

I will try myself something like that, I don't know if within 2 years or in 10... in this very site, if I can. But right now I have other things to do.

But on the meanwhile, I just share this idea.

Have a nice day.

Updated

The thing about AI is that you can just slap a Post-It note on any picture reading "Mona Lisa" and the AI will believe that's the Mona Lisa.

lonelylupine said:
The thing about AI is that you can just slap a Post-It note on any picture reading "Mona Lisa" and the AI will believe that's the Mona Lisa.

In what I (nebulosely) think, the criteria of what is "good" continues to be given by humans.

The most common kind of IA system, only "regresses" or is "trained", given "examples" or "cases"... in which the level of goodness, score or similarity, is already given.

As an initial approach, I do not think that an analysis of the image itself is even necessary... although that may be the advanced case, later.

In reality, I am not very willing to talk much about the theme right now... the first thing that I have to do, is to regain some skills that I have neglected for decades. But that I have already begun.

One thing is to just "give and idea", and the other, is to put the "sweat and courage" to implement it.

But I believe, that once given the idea, probably you will see it implemented, soon, even if not by me.

=).

Regards.

binaryfloof said:
There was an attempt at this once, over at topic #25785.

Very interesting @binaryflood, seems that efectively you had 8 months ago an idea of rewriting an application, that will include a module of post recommendation based in algorithms relating to the philosphy of Machine Learning... and that you are inviting people to your team to continue develop such application.

I can only congratulate you and desire that your efforts will lead to great success.

I myself, just beginning to have vague ideas about the subject.

Guess that what I have to do is to try your application, and maybe in the future we could exchage ideas.

=).

Congratulations again.

  • 1