HN Explorer

Ask HN: Review my startup, a search engine to find what you like

11 points by jey ↗ HN
We're building a search engine to "find what you like", and are targeting t-shirts as the first application. The site works by showing you two shirts at a time and asking you to choose the one you like more, and uses your votes to learn your preferences and find shirts you'll probably like.

Right now the site is pretty bare-bones and the "suggestions" at the bottom of the page are just randomly generated, but the site will work by using machine learning algorithms to compare your votes to other people's votes and come up with predictions of what you'll like.

We'd love it if HN could help us out by providing feedback/comments/criticism and the initial dataset of votes for the algorithm to analyze.

Thanks!

9 comments

[ 4.0 ms ] story [ 49.5 ms ] thread
tgrass
1) I'd give the suggestion box more real estate. That's the site's focus.

2) If your using an ELO style algorithm, why not provide a "Neither" option?

3)It doesn't seem very effective in ranking the style of the shirt...Will it differentiate between choices of color, style and content?

tgrass
Sorry, just reread and saw the suggestions are random for now. Disregard #3
jey
> 1) I'd give the suggestion box more real estate. That's the site's focus.

Agreed.

> 2) If your using an ELO style algorithm, why not provide a "Neither" option?

I've done some testing with a "neither" button and people tend to over-use it. It's definitely an oft-requested feature though, so it would be nice to find some way of providing a neither option while still discouraging overuse.

Also, it's not quite ELO-style. This is to provide personalized recommendations, not a global ranking.

acgourley
Can you give some instant reward to users? Sure in 20 clicks they are getting a set of things they should buy, but what about people with a shorter attention span? What if you could see statistics about how your opinion lines up with other people? What % of people agree with you, perhaps the trending popularity of the shirt over time.
apowell
1) I clicked a ton of shirts; does it ever end? Where are the suggestions?

2) I clicked "help" and got nothing. Chrome, Safari, and Firefox, all on a Mac.

3) Could you preload the next comparison so the site is snappier? If the next comparison depends on the results of the previous comparison, perhaps load the comparison tree up 2-3 levels deep. The experience would be much nicer if it everything was totally instant.

jey
> 1) I clicked a ton of shirts; does it ever end? Where are the suggestions?

The suggestions are at the bottom. I'll make some changes to make this attract more attention and to make its purpose clearer. But no, there is no "end" per se.

> 2) I clicked "help" and got nothing. Chrome, Safari, and Firefox, all on a Mac.

Oops, fixed!

> 3) Could you preload the next comparison so the site is snappier?

Good suggestion.

astrofinch
I think it would be entertaining to read an estimated percentage chance that I'd buy each of the shirts you showed me.
karlclement
Hello there,

If you would like to drive additional traffic to your application, I would recommend integrating a Facebook application. People browsing there Facebook News Feed enjoy seeing the responses of others. Maybe you could update their status with some statistics. You could also make users invite their Facebook friends to try out the application.

You could also show what their friends answered on the application page, if they have signed in using their Facebook account.

This could also be a cool feature, you could classify the person by a certain characteristic. Example, depending on the users answers, you could display the users personality. They could be categorized as Artistic, Logical, Funny, Nerdy etc. You could also change the background style of the application depending on their personality.

These are just a few ideas that came to mind,

Good luck!

Karl