Jan 7, 2011 15:30 GMT  ·  By

Gmail's Priority Inbox, which tries to determine which emails are important to the user and which can be pushed to the side, has proven quite successful, when considering the scale of the problem. Google already revealed that users with Priority Inbox enabled spent less in Gmail and less time reading unimportant emails.

Now, several Google employees have published a paper detailing the issues Priority Inbox tries to tackle and what they've done to customize the algorithm to each user.

"The challenges include inferring the importance of mail without explicit user labelling; finding learning methods that deal with non-stationary and noisy training data; constructing models that reduce training data requirements; storing and processing terabytes of per-user feature data; and finally, predicting in a distributed and fault tolerant way," Douglas Aberdeen, Ondrej Pacovsky and Andrew Slater from Google Zurich explained in the paper (PDF).

Gmail already filters some emails, those marked as spam never hit the inbox. But, while most people would agree on what is spam and what isn't, they would deem 'important' different types of emails.

One of the biggest challenges was designing an algorithm that is good enough that it works for everyone out of the box, as soon as users enable the feature.

Beyond that, Priority Inbox becomes more accurate the more you use it. The algorithm looks at several indicators, like how fast a user interacts with a certain message, replies and of course, the labels indicating whether the email is important or not.

The research paper goes over some of the theory and maths behind the algorithm which could prove interesting for those tackling similar problems.

The regular users should be more interested in the end results. Based on the data gathered from the way Googlers used Gmail, with or without Priority Inbox, it seems like the feature is making an impact. Priority Inbox users spent on average 6 percent less time reading email and 13 percent less reading unimportant email. [via RWW]