Hey, Samuel (et al.), I just got done responding to yet another lamentation on Hacker News about the decline of RSS and blogging, and it occurs to me:
It’s hard (-er than it could be) to discover blogs based on my interests. I thought to share my OPML with the world and then I thought “what if we all did that, and we had some software that would scan all our subscriptions and suggest others that are popular, and maybe also suggest feeds that I might be missing based on OPMLs from other folks who have significant overlap with my OPML?”
Or something.
Maybe NewsBlur could do that? Maybe it already does and I just don’t know?
Social feeds, like del.icio.us had social bookmarking.
However, I would consider the list of sites a user subscribes to as sensitive information, at the same level as their browser history.
If this feature were to be implemented, it is important that it’s done in such a way that it cannot reveal the feeds that an individual user subscribes to. A potential issue could be that, if a Eve knows a big subset of the feeds that Bob subscribes to (for example, by following Bob’s Blurblog), then Eve could create a new Newsblur account, add those feeds, and then deduce which other feeds Bob subscribes to from the social recommendations given by Newsblur.
One way of implementing this feature in a privacy-preserving way is that, when recommending sites to Eve, the algorithm uses all the sites Eve is subscribed to, but only those sites which Bob has shared to their Blurblog. Conversely, recommendations for Bob are based on Bob’s whole list of feeds, but only on those sites that Eve has shared to her Blurblog.
Another way is to only make recommendations based on pairs of sites (e.g. subscribers to Site B are more likely than average to also be subscribed to Site A), and do so only where a minimal number of users are subscribed to both feeds (for example, only recommend a Site A to Eve based on her subscription to Site B if there are at least 10 users subscribed to both A and B).
Insightful comments and thanks for writing these up. I agree that privacy must be preserved and no feed will ever be recommended unless it has a k-anonymity factor of > 2. That means that a feed has to have at least two pathways (users subscribed to them with relations to the feed you’re looking at) in order to be recommended. That way nobody is individually identifiable. There are other privacy preserving methods that I could use, such as L-diversity.
As for how this is implemented, I’m thinking about a new dialog box with recommendations for individual stories/topics, feeds, and users to follow. There’s a lot to build, both backend and frontend, but I recently hired a backend developer to help me make NewsBlur easier for other developers to install and contribute to. At that point, hopefully in a month or two, I’ll have what I need to hire more developers to work on that code for this feature. It’s a slow process but it’s being actively worked on.