AI in Mozilla

Mozillla 1.3b features a machine learning algorithm for the url bar auto-complete and a data collection mechanism to help the authors, some Stanford folk, refine it:

This release of Mozilla marks the kickoff of a research project to apply machine learning to improve the browser's autocomplete feature. The project's first phase needs as many Mozilla users as possible to participate in collecting data. All you need to do is turn on a pref, use this build for daily browsing for a couple of weeks, zip up a data file on your local drive, and submit it online.

It's great to see this type of work in Mozilla and the use of an opt-in data collection methodology. On the other hand, recency and frequency have been shown in the HCI literature to be the best predictors of revisitation. The bug that tracked the latest overhaul on this feature in Mozilla was far from a rationale and informed discussion. If it takes a Stanford professor getting involved to get this right, so be it. But it shouldn't.

Personally, my biggest problem with autocomplete is the lack of flexible slicing/dicing of the url structure. If I autocomplete to a.com/b/c/d, I want to easily be able to get to /b/ without the delete key. |||
Posted at NaN:NaN, Published in: Mozilla