Taste graphs will transform fashion
blog.chicisimo.com
blog.chicisimo.com
You're welcome to make a startup based on that idea ;) up here in Canada where we have real seasons I'll find you some great test users!
(Although I've been to the Bay Area and it's not without weather challenges)
It would be better to train a classifier to learn what clothes fit the current trend.
https://medium.com/@Pinterest_Engineering/taste-graph-part-1...
How is this different/better than what they are doing?
Just a heads up.
Not that it matters, as I'm certain other companies use more modern techniques (deep learning) for "outfit advice".
Bear in mind that HN began as a community of startup founders and enthusiasts, and even though the audience has broadened, it's still part of the culture to be charitable and supportive to people who are just starting out and trying to get something off the ground.
Not everyone entering the startup world has an MIT PhD (we hope!) and the ability to write content that has both technical depth and literary polish.
We all have to start somewhere, and the only way to start getting better at something is to put something out there and see what comes back.
Critique is fine, but we can be constructive and encouraging to help people improve - just as we'd hope people would to for us if we're taking a risk on trying something new.
- We do use DL. I didn't mention it because it is not relevant in the context of the post;
- How we use DL. One of the jobs of the graph is to tell the DL algorithm what content needs what type of descriptors. The graph can do this thanks to the different levels in our ontology, and because it understands our content in its context: it understands people's interaction and tagging. DL is a small part in our entire infrastructure;
- The DL market. Lots of companies use DL to identify attributes in an image, and the level they achieve is impressive. Having had long discussions with the best of these companies, I can tell you that building the correct ontology is nearly impossible without the entire infrastructure. We are happy building the intelligence that tells DL what to do, and then attaches descriptors correctly to outfits and taste profiles;
- Our patents. They cover a few relevant aspects in the online fashion market: a system to tag fashion images with shoppable products; a system and method to capture/understand how people mix and match clothes in outfits and closets; and more.