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

I had an idea about this recently, and would love for someone to further develop it. I already implemented a proof of concept, but won't have time to build it to a complete product any time soon.

The (simplified) idea for recommending movies was:

1. Crawl a torrent site (say, the pirate bay) for as many torrents you can find. Extract the title, description, magnet link and category for each torrent. 2. Use the DHT network to find out which IP addresses are downloading what torrents 3. Use these (torrent, ip) tuples to build a recommendation based on "People who download X also download Y"

This way you can try to find movies that are liked (or actually, downloaded) by people with similar taste to you. The quality of this recommender is quite impressive in my opinion.

Email me at arno at vanlumig dot com if you want to see the proof of concept, I won't post a link here because the performance is quite bad and I'm sure the server won't be able to handle any significant load.

I will make the data (torrent metadata and data about who downloaded what) available as soon as I've anonymised the IP addresses, also mail me if you're interested in that data.




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

Search: