Applying touch gesture to improve application accessing speed on mobile devices.

Abstract

The touch gesture shortcut is one of the most significant contributions to Human-Computer Interaction (HCI). It is used in many fields: e.g., performing web browsing tasks (i.e., moving to the next page, adding bookmarks, etc.) on a smartphone, manipulating a virtual object on a tabletop device and communicating between two touch screen devices. Compared with the traditional Graphic User Interface (GUI), the touch gesture shortcut is more efficient, more natural, it is intuitive and easier to use. With the rapid development of smartphone technology, an increasing number of data items are showing up in users’ mobile devices, such as contacts, installed apps and photos. As a result, it has become troublesome to find a target item on a mobile device with traditional GUI. For example, to find a target app, sliding and browsing through several screens is a necessity. This thesis addresses this challenge by proposing two alternative methods of using a touch gesture shortcut to find a target item (an app, as an example) in a mobile device. Current touch gesture shortcut methods either employ a universal built-in system- defined shortcut template or a gesture-item set, which is defined by users before using the device. In either case, the users need to learn/define first and then recall and draw the gesture to reach the target item according to the template/predefined set. Evidence has shown that compared with GUI, the touch gesture shortcut has an advantage when performing several types of tasks e.g., text editing, picture drawing, audio control, etc. but it is unknown whether it is quicker or more effective than the traditional GUI for finding target apps. This thesis first conducts an exploratory study to understand user memorisation of their Personalized Gesture Shortcuts (PGS) for 15 frequently used mobile apps. An experiment will then be conducted to investigate (1) the users’ recall accuracy on the PGS for finding both frequently and infrequently used target apps, (2) and the speed by which users are able to access the target apps relative to GUI. The results show that the PGS produced a clear speed advantage (1.3s faster on average) over the traditional GUI, while there was an approximate 20% failure rate due to unsuccessful recall on the PGS. To address the unsuccessful recall problem, this thesis explores ways of developing a new interactive approach based on the touch gesture shortcut but without requiring recall or having to be predefined before use. It has been named the Intelligent Launcher in this thesis, and it predicts and launches any intended target app from an unconstrained gesture drawn by the user. To explore how to achieve this, this thesis conducted a third experiment to investigate the relationship between the reasons underlying the user’s gesture creation and the gesture shape (handwriting, non-handwriting or abstract) they used as their shortcut. According to the results, unlike the existing approaches, the thesis proposes that the launcher should predict the users’ intended app from three types of gestures. First, the non-handwriting gestures via the visual similarity between it and the app’s icon; second, the handwriting gestures via the app’s library name plus functionality; and third, the abstract gestures via the app’s usage history. In light of these findings mentioned above, we designed and developed the Intelligent Launcher, which is based on the assumptions drawn from the empirical data. This thesis introduces the interaction, the architecture and the technical details of the launcher. How to use the data from the third experiment to improve the predictions based on a machine learning method, i.e., the Markov Model, is described in this thesis. An evaluation experiment, shows that the Intelligent Launcher has achieved user satisfaction with a prediction accuracy of 96%. As of now, it is still difficult to know which type of gesture a user tends to use. Therefore, a fourth experiment, which focused on exploring the factors that influence the choice of touch gesture shortcut type for accessing a target app is also conducted in this thesis. The results of the experiment show that (1) those who preferred a name-based method used it more consistently and used more letter gestures compared with those who preferred the other three methods; (2) those who preferred the keyword app search method created more letter gestures than other types; (3) those who preferred an iOS system created more drawing gestures than other types; (4) letter gestures were more often used for the apps that were used frequently, whereas drawing gestures were more often used for the apps that were used infrequently; (5) the participants tended to use the same creation method as the preferred method on different days of the experiment. This thesis contributes to the body of Human-Computer Interaction knowledge. It proposes two alternative methods which are more efficient and flexible for finding a target item among a large number of items. The PGS method has been confirmed as being effective and has a clear speed advantage. The Intelligent Launcher has been developed and it demonstrates a novel way of predicting a target item via the gesture user’s drawing. The findings concerning the relationship between the user’s choice of gesture for the shortcut and some of the individual factors have informed the design of a more flexible touch gesture shortcut interface for ”target item finding” tasks. When searching for different types of data items, the Intelligent Launcher is a prototype for finding target apps since the variety in visual appearance of an app and its functionality make it more difficult to predict than other targets, such as a standard phone setting, a contact or a website. However, we believe that the ideas that have been presented in this thesis can be further extended to other types of items, such as videos or photos in a Photo Library, places on a map or clothes in an online store. What is more, this study also leads the way in tackling the advantage of a machine learning method in touch gesture shortcut interactions.