by: Yang Li
Authors
Yang Li is a senior research scientist working for Google, and used to do research in computer science and engineering at the University of Washington.
Presentation Venue
"UIST '10 Proceedings of the 23nd annual ACM symposium on User interface software and technology" as per the document specs from the ACM digital library, dl.acm.org. The presentation took place in New York City, NY, 2010. (Yep. Again)
Summary
Gesture Search uses gesture inputs to search data on mobile devices. The idea stemmed from a desire to combine the convenience of a "Search All" option with the ease of gesture input on a mobile device. The idea is that by using a learning algorithm and short gestures, it would be possible to greatly enhance the speed and experience of users interacting with their mobile devices.
Hypothesis
A couple of different hypotheses were investigated. The obvious was that Li supposed he would be able to implement a design which allowed users to search the data in their phones with the ease of gestures. He also had to determine if it was possible to distinguish with relatively high accuracy whether an input was intended as a GUI touch or a gesture. He also hypothesized that the average length of queries a user would need to use would vary based on the volume and complexity of data on the mobile device.
Methods
The search was implemented on Android devices with decent success. It was made available by download to a large audience for testing, and after a month, data was collected to determine the frequency of use, the accuracy of gestures, the average gestures required, and the types of information most frequently sought.
Results
After successfully implementing Gesture Search, the data from initial test users was collected and showed that 66% of all gesture searches were made for contacts, and average query length needed as time progressed stayed consistently rather small (less than 5 letters, 80% less than 2). This was especially true as the optimization allowed frequent queries to be returned sooner so individual gestures were more closely associated with specific searches.
Discussion
I really liked this idea. In the first couple of paragraphs where it was talking about gestures, I was already thinking: "Great. Another attempt to make us learn lots of vague gestures that somehow map to some application that I may or may not have on my phone. Why doesn't someone just implement the ability to stroke out a letter by hand, so we don't have to click through lots of interfaces to find our contacts or applications?" A couple of paragraphs later, I was grinning broadly at the realization that the paper had read my mind and altered its content for me! Even if this new application didn't receive any positive feedback from testing users, it'd still be a success in my mind due to my own conceptual bias for the idea. I was worried about conflicts with ordinary GUI touch interactions, but the application addresses those - albeit I cannot be sure how robust it is. I am curious to know how the gesture recognition can learn the user better. I understand that some machine learning could be applied, and that handwriting recognition software isn't all THAT bad, but do our mobile devices come equipped with the space and computation power to continuously learn from its users, or will this cause the phone to crash in 3 months of regular use? That's the sort of thing that would have me waiting 6 months to see what others think about it. I find the idea of using gestures that we already recognize ( i.e. letters of the alphabet and similar symbols ) severely more appetizing than mapping abstract gestures that people can create on their own. Perhaps this will lead people to the desire to create their own gestures, and then later it'll be more feasible. But, for now I feel like branching out into the new frontier should have some explicit ties back to home to help people get on board.
No comments:
Post a Comment