- Twitter: Jan 11th was the highest usage day ever (source: @ev via techcrunch)
- Tweetdeck: did 4,143,687 updates on Jan 8, yep 4m. Or, 48 per second (source: Iain Dodsworth / tweetdeck internal data)
- bit.ly: last week was the largest week ever for clicks on bit.ly links. 564m were clicked on in total. On the Jan 6th there were a record of 98m decodes. 1100 clicks every second.
Because Twitter is (mostly) public, and most tweets can be accesssed by API, it’s also an amazing platform for analyzing sentiment on a wide range of topics. 140 characters isn’t much of a payload, but when you include analysis of included links, and bundle in public data about individual Twitter users (also available by API), a smart developer can construct an almost limitless array of lenses into the global zeitgeist.
One of our portfolio companies, AppStoreHQ, has been working this problem for a while now. The company is focused on iPhone app discovery – with 120,000 apps in the App Store, it’s gotten hard for consumers to find what they’re looking for, and even harder for iPhone developers to get the attention of potential customers. AppStoreHQ started by indexing leading blogs to surface the most-talked-about apps among the “official” voices in tech. Then they added a similar feature built on top of the Twitter API, offering a rolled-up view of the apps currently seeing the most tweet velocity.
Over the holidays I was talking with the AppStoreHQ team about the “holy grail” of app recommendations: providing personalized suggestions based on the apps you use and love the most. With the sole exception of the iTunes Genius feature, all of the existing offerings require software downloads, website registration, out-of-workflow “voting” actions and other sources of behavioral friction that severely retard adoption and scale. Our question was: how could you offer a scalable social recommendations service for iPhone apps with the least possible friction?
As often happens, we already had the answer, we just hadn’t realized it. Our Twitter Hottest feature is built by rolling up and analyzing individual tweets about iPhone apps. Why couldn’t we “flatten out” that data set and instead look for similarities between individual apps and people? We ran a few quick database queries and found more than 10,000 people who had tweeted about an individual iPhone app at least once, and more than 100,000 app-related tweets. When we combined that with our own dataset from thousands of AppStoreHQ members sharing, bookmarking and buying apps, we realized we had amassed the largest social dataset about iPhone apps outside iTunes itself.
It took some serious heads-down effort to make it so (Ian has a great post up on the technical and scaling challenges of building social recommendations in Ruby), but this week we shipped our first rev of a social recommendations service for iPhone apps with the simplest and lowest-friction user model we could come up with: just tweet about the apps you love and we’ll automagically build out a set of recommendations for you. The more you tweet about apps, the more accurate these recommendations become. And you don’t even have to register at AppStoreHQ to get your recommendations – just visit the page we create for you based on your Twitter handle (e.g., http://www.appstorehq.com/users/yourtwitterhandle ).
You can learn more about this new service on the AppStoreHQ blog (and in today’s TechCrunch writeup). I’m excited about the service itself, but the project also opened my eyes to the broader possibilities available by applying the same pattern in other domains. It wouldn’t be hard to build a Twitter-based social recommendations service for almost any category of media – websites, books, movies, etc. – as long as you have a set of “canonical” URLs to map against the links included in the Twitterstream.
Are there other examples of services like this out there already? Let me know – I’d love to hear about them.