Hacker News new | past | comments | ask | show | jobs | submit login

> I would guess that, apart from the immense effort of building it, delivering personalised search results like this would be enormously expensive in storage for the search engine.

How expensive would it really be?

You have O(the_internet) in pages and metadata, and you have O(world_population) in user preferences. So long as your index structure allows those to be mostly decoupled (if I had to take a first crack at it I'd probably try to embed preferences and pages into a vector space and build a projection index -- exact matches are hard in that system, but decent personalized results are easy), I don't think it'd be all that much more space than a non-personalized search engine, especially given that the world population is kind of small compared to the size of the internet.

For that matter, the web isn't thaaat big (ignoring images and video). The entire common crawl can fit on a single $3k-$5k disk uncompressed.




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: