I’ve been waiting for Google Reader to roll out the feed recommendation service that just popped into my feed reader yesterday. I think this service will actually be far more useful for people who consume a ton of feeds than it will for those who are casual feed readers. This assertion isn’t based on any facts, it’s just a hunch that those who already read a lot of feeds are generally interested in reading even more feeds. Making previously-unknown feeds available to that audience (and I put myself in that audience) will likely lead to that audience consuming even more feeds.
Even though the first version of this feature is very rudimentary, it looks like Google has the 4 things you really need to build a great feed collaborative filtering product:
Large number of feed readers – Having a large number of readers does two things. Assuming the readers are not completely homogeneous, you get a wide variety of feeds to use as inputs into your collaborative filtering engine. Second, having a large audience of feed readers gives you a bigger pool of people over whom to make inferences about what feeds a user will like based on what the community is doing.
Good data on feed subscribers – The simplest way to divine the popularity of a given feed is to look at how many subscribers it has. Google owns Feedburner, which probably has more data about feed subscribers than any other single service on the web. Using subscriber numbers as a really rough proxy for popularity ought to be enough to get you started.
Attention data on which feeds and items are read and shared – The trends tab in Google Reader exposes how I interact with the feeds to which I subscribe. How often they post, how often I read them, what percentage of feeds to which I subscribe I subsequently ignore, etc. This is the kind of attention data you need to combine with the good data on feed subscribers to figure out how popular a given feed is in practice. There are some other relevant indicators, such as trackbacks, comments, and the like that also give you some sense for user engagement with the content.
Computing horsepower to perform collaborative filtering – It takes some good computer science and some good hardware to do the number crunching and algorithm development to make a good collaborative filtering solution work. Anyone question whether Google has both of these things?
I’m glad to see that Google released this product. As a former Googler myself, I think these are the kinds of products that will get Google to be more social. The way forward is likely in what I would describe as “passively social” products that leverage data in the background without explicit user action (tagging, labeling, voting, etc) as opposed to “actively social” products that require a lot of explicit action by users.
Blogged with Flock