What to Read Next : InstaRank - v1.0

One of our ambitions since taking on Instapaper has been to build useful ways to make the reading experience more productive.  To us, that means making it easy to find the best article to read (or video to view) for the moment you’re in.  To that end, we’ve added a bunch of sorting and filtering capabilities into the new iOS 7 app for iPhones.  (They’ll show up in the very near future on our iPad and Android apps, and in the web interface.)

It has long been evident that users want more and better ways to find particular articles in their “Read Later” queues.  It can be pretty unsatisfying to scroll endlessly through items in reverse chronological order.  Some highly useful sorting and ranking can be done in a straightforward way, such as ordering your articles by article length or date.  But the power of data and algorithms can enable much more interesting, useful, and personalized tools for ranking.

The newly released Instapaper 5.0 features InstaRank, Instapaper’s first algorithmic ranking and sorting system for your saved links.

A web link’s InstaRank is determined by the following factors:

• The number of overall saves/reads/likes on the link.

• The number of saves on the link in the last 4 hours, indicating trending nature.

• The age of the link, since it was first seen in the Instapaper world.

• The popularity of the link within its domain, meaning the number of saves/reads/likes on the link compared to the domain average.

• The popularity of the domain compared to other domains in the Instapaper world, meaning the domain average saves/reads/likes compared to other domains in the last 2 weeks.

• Whether we see a link from some lesser known domain that receives surprising levels of attention, measured by saves/reads/likes.

To fully shape InstaRank, we explored questions like “Did many web pages of this domain get likes, or did just one article get many hits?” and “Have links of this domain performed well historically, measured in terms of likes, reads, saves?”. We found users prefer some web domains or sources of information more than others, and used visualizations like the one shown below (which shows the top 18 domains that Instapaper users most interacted with on Sept. 4th, 2013) to identify what domains might generate popular links.

We also knew that the buzz around older articles decays with time and newer articles start gaining interest, but found this pattern to be less straightforward than we hoped. As it turns out, over 66% of Instapaper users read/like/share an article approximately 3-5 days after they bookmark it, which causes the number of reads/likes on an article to rise significantly beyond the 4th day or in the first weekend after it is bookmarked. We needed to adjust InstaRank’s popularity decay function to suit this user behavior.

Try InstaRank for yourself through the Popularity sort in your “Read Later” list on the iPhone. We would love to hear your feedback and suggestions on features to incorporate (or scrap) when ranking and whether it’s helping you sort, explore and discover great content to read next on Instapaper.

And keep an eye out for future versions, which will leverage social trending scores and categorical classification of links.

– Suman Deb Roy, @_RoySD