Has anyone created an RSS-based recommendation engine yet?
11 comments
SeekSift (http://www.seeksift.com/) is headed in that direction, though it works a little differently from what you described.
I hadn't thought about the possibility of producing OPML as the output, but it's certainly something I could add in a future version (it's still in a fairly early stage of development right now).
I hadn't thought about the possibility of producing OPML as the output, but it's certainly something I could add in a future version (it's still in a fairly early stage of development right now).
If there's a way of uploading an OPML file on your site, I'm not seeing it.
Most people who are fanatic about RSS aren't going to be able to manually input the URL to each of their feeds due to the sheer number of feeds they subscribe to.
Most people who are fanatic about RSS aren't going to be able to manually input the URL to each of their feeds due to the sheer number of feeds they subscribe to.
You're right, there's no support for OPML at all.
There's a new design in the works for adding feeds, different from what's there now, but I hadn't thought of using OPML until you mentioned it.
I'll post it in the next few days, so check back later in the week if you get a chance.
Alternatively, if you have any other feedback, let me know here, or just email me (my address is in my profile).
There's a new design in the works for adding feeds, different from what's there now, but I hadn't thought of using OPML until you mentioned it.
I'll post it in the next few days, so check back later in the week if you get a chance.
Alternatively, if you have any other feedback, let me know here, or just email me (my address is in my profile).
I'm not sure if you're still following this thread, but I just wanted to tell you (and I didn't see any contact information in your profile, otherwise I'd have just contacted you directly) that you can now upload an OPML file when adding feed sources.
Other changes are in the works, too; send me an email or reply to this thread if you'd like to be kept un-to-date on those as they happen.
Other changes are in the works, too; send me an email or reply to this thread if you'd like to be kept un-to-date on those as they happen.
I'll check it out, thanks.
I'm not sure why the format is important.
Building recommendation engines is really hard from an algorithmic perspective. How you format the inputs and outputs seems negligible in importance.
Building recommendation engines is really hard from an algorithmic perspective. How you format the inputs and outputs seems negligible in importance.
It's not about the format, it's about what being in someone's OPML file means. It means that someone finds the information from a specific site to be so useful that they want to receive every update from the site so they can at least take a glance at it. Votes on specific links are a far less valuable indicator of reputation. No one wants to see everything that's submitted to Scribd; a few things just happen to get voted to the top of social networking sites because they're funny or interesting. That doesn't say anything meaningful about Scribd's reputation.
Google has a huge advantage in this area since they collect which items from an RSS feed people actually read on Google Reader. They know that even though I subscribe to the del.icio.us/popular and digg RSS feeds, I rarely read any of the items, whereas I read almost everything from dive into mark or Jeremy Zawodny's blog.
As far as the output is concerned, it doesn't have to be traditional "output" at all. It can be an increased PageRank for a site, or a Firefox extension that shows a reputation meter for every site you visit.
The format isn't important, but the things people subscribe to are very important.
Google has a huge advantage in this area since they collect which items from an RSS feed people actually read on Google Reader. They know that even though I subscribe to the del.icio.us/popular and digg RSS feeds, I rarely read any of the items, whereas I read almost everything from dive into mark or Jeremy Zawodny's blog.
As far as the output is concerned, it doesn't have to be traditional "output" at all. It can be an increased PageRank for a site, or a Firefox extension that shows a reputation meter for every site you visit.
The format isn't important, but the things people subscribe to are very important.
The format of the output is somewhat less important, although I just seem to think RSS feeds are a inherently superior way to get content than a list of links.
The INPUT on the other hand is extremely important. Take a look at Reddit... you have to manually up/down vote articles to make the recommendation system go. Not only is that tedious, but I don't believe people actually have a good grasp of the content they are looking for.
RSS feeds, on the other hand, are much more relevant. And getting a list of someone's feeds requires no work on their part to describe to you their interests.
Even better, online readers/aggregators such as Google Reader have an even more extreme advantage in being able to tell what you read and how often you read it (as the other poster pointed out) more accurately than your OPML file. It's a gold mine for anyone trying to build content recommendation systems.
Take a look at Reddit... it's become a collective RSS feed aggregator plus stupid pictures of cats. If the recommendation engine worked, it would be a GOOD collective RSS feed aggregator. I believe you can get there by working with RSS feeds and actual reading statistics as inputs rather than manual link clicking.
The INPUT on the other hand is extremely important. Take a look at Reddit... you have to manually up/down vote articles to make the recommendation system go. Not only is that tedious, but I don't believe people actually have a good grasp of the content they are looking for.
RSS feeds, on the other hand, are much more relevant. And getting a list of someone's feeds requires no work on their part to describe to you their interests.
Even better, online readers/aggregators such as Google Reader have an even more extreme advantage in being able to tell what you read and how often you read it (as the other poster pointed out) more accurately than your OPML file. It's a gold mine for anyone trying to build content recommendation systems.
Take a look at Reddit... it's become a collective RSS feed aggregator plus stupid pictures of cats. If the recommendation engine worked, it would be a GOOD collective RSS feed aggregator. I believe you can get there by working with RSS feeds and actual reading statistics as inputs rather than manual link clicking.
Google does have a recommendation ("You might also like") system for when you add a feed to your iGoogle page.
We're obviously all going to be using RSS some day so this is inevitable. I'm just wondering if someone's done it yet.