[dvirsky]

There were a few attempts at that in the past, one being http://omgili.com/ that now seems to return pretty much garbage.

BTW About 12 years ago I was building this search engine, and I was toying with the idea of building a classifier that classifies web pages based on their "genre" rather than category, so you can limit your search for shopping websites, forums, blogs, news sites, social media, etc. It was a bitch to train, and my classifier's algorithm was pretty crappy, but it showed some potential.

I think today modern search engine do that behind the scene, and try to diversify the result to include pages from multiple genres, but they usually don't let you choose.

[dood]

Heh, classifying by "genre" is exactly what I was thinking of doing.

Had some debate with myself if I should start by focusing on training for shopping pages (product pages & product reviews) - because that might make some money; or start by training for forums - which I'd enjoy a lot more. Or build a more general system which would definitely never work and never get finished.

Google actually let you filter by "discussions" until a few years ago, so they certainly do this kind of classification. It didn't work perfectly but sometimes did the trick. Don't know why they removed that feature.

[petra]

Google removed it because they aim at the mass market.

Another perspective: people who find answers in forums are less likely to be interested in ads. And who knows, maybe making search shitty(in so many ways, not just formus), ad revenues rise ?

[dvirsky]

IIRC I found that the easiest was to train on shops, forums, and porn. But another tricky bit was conceptual - genre and category overlap sometimes (e.g. porn). Anyway I couldn't get it to yield proper results. But today we have things we didn't back then like opengraph and schema.org tags, that give more semantic info.

[jacquesm]

Iirc Gigablast had such a feature.