Is it possible for NewsBlur to suggest people to follow based on the similarities of ‘intelligence trainer’ and shared items?
Yeah, this is one of those features that was suggested post-launch and I absolutely love it. But it’s a lot of work. I want to recommend users based on a few criteria:
- Who they follow intersected with who you follow (similar interests, sort by size of intersection)
- Who follows them intersected with who you follow (friends of friends)
- Similar subscriptions. This is tough because technically that data is private, even though I could mine it to find similarities. I’m leaning against doing this, because even surfacing some latent data based on your private data is ethically dubious.
- Popular users. Easier to calculate (# of stories * weight + # of followers * weight), but can be somewhat biased, as popular users only get more popular due to this formula and the fact that they are recommended consistently. See Twitter’s SUL for more debate.
So take 1, 2, and 4 and you’ll see what I’m going to code sometime this week (if not today!).
Similar subscriptions. This is tough because technically that data is private, even though I could mine it to find similarities. I’m leaning against doing this, because even surfacing some latent data based on your private data is ethically dubious.
What if this privacy setting is left to the user? If enabled, their data will be used to suggest who to follow.
For example, I love to read comics and would want to see other NewsBlur users who have same interest and what they are sharing. Perhaps I will stumble upon another great resource through them.
In that case, why not surface the data based on the feed, not the user. I have no issue with showing all extractable data from a feed, even if only a single user is following it. (The exception being that I don’t autocomplete sites with fewer than 2 users in the Add Site dialog. I also check for private-esque keywords, including authentication parameters.)
So if you want to find more comics feeds, you’d do well with finding similar sites based off a specific site, rather than the users who follow that site. There are algorithms for figuring out which sites are similar, but that’s a whole heaping of work. I’m planning to do it, but only after I do the next big three things on my list:
- Launch social
- Rewrite River of News and read stories
- Write iPad app and update both iOS apps for social
That’s at least 6 months of work. But this is a great idea and it’ll happen soon enough.
It is not just about getting links to new feeds. It is about getting relevant and interesting content. It’s not like each and every item of a feed would be appealing to a user.
Take daringfireball.net for example. His number of posts per week is 48.8. Lets say the site is suggested to me by NewsBlur through an algorithm to figure out similar sites.
I add it. But would I be able to keep up with this super busy feed? I don’t think so.
On the other hand, Newsblur suggests me to follow X, Y and Z users, because these guys are subscribed to daringfireball.net and some similar sites, just like I am.
I can go to their BlurBlog and see what they are sharing. They would have scanned the feed, partly or completely, and shared interesting, useful and share worthy items.
As our subscriptions list is similar, those shared items would be in my feed list too. But they could be hidden under numerous other items. I might have ignored it.
I think that is why we visit news aggregators like Hacker News. We want to see content that is of appeal to us. We can subscribe to hundreds of site but that would be ungainly.
On the other hand, news aggregators sites provide us with popular content sorted by a community of like minded people.
I want to know users whose interests are similar to mine - a community of like minded users of NewsBlur. This would increase my chances of getting useful content with minimal efforts.
If 1,2,4 is good enough to introduce me to them, great! Otherwise I am willing to sacrifice my privacy and get it used for suggestions.