| Text entry is one of the most basic tasks in Human-Computer-Interaction(HCI),which is a major approach for users to express their interaction intentions to the computer Beyond the conventional physical keyboards,new interaction interfaces(e.g.touch-screens)are emerging day by day,which increases the naturalness of interaction,but significantly decreases the interaction performance.This is due to a higher level of input noise and the limitation in the feedback channel,and as a result,a low signal-to-noise ratio of the input signal.How to balance the tradeoff between the naturalness and efficiency of interaction has become a challenging problem in the field of HCI.Researchers tend to use the B ayesian method to solve this problem,but they usually consider only the information of touch point location,which limits the functionality of this method.In this thesis,focusing on various input interfaces,we optimize the basic Bayesian method in text entry by increasing the precision of touch model(increasing the calculation precision),gener-alizing the computation for uncertain mapping in the input(generalizing the computation method)and modeling the hand movement during typing(increase prior knowledge)The corresponding techniques significantly increase the performance of natural text entry We also propose the calculation of word clarity,and offer optimization metric as well as public-available datasets for obtaining appropriate task phrase sets,which serve as a key tool in text entry evaluation studies.Specifically,this thesis presents four research contributions:(1)Proposing the precision-optimized touch model for touchscreen soft keyboards:Focusing on the fat finger problem,we design and implement a high-precision technique for inferring touch point locations.Based on this technique,we quantify the correlation between interface size and input speed,accuracy,as well as touch point distribution.The results improve the goodness of fit between the input behavior and the touch model in the B ayesian method for soft keyboards.The algorithm can achieve 35 WPM on smartwatch QWERTY keyboards,which rivals the input speed on smartphones.The algorithm can also decrease the input error rate on smartphones by 23%(2)Proposing the generalized Bayesian method for uncertain mapping in the input:Focusing on the limitation of the basic B ayesian method that it can only be applied to cer-tain mapping in the input,we propose a generalized computation method for incorporating multiple kinds of input error and multiple input signal channels.This allows us to develop the first multi-cursor text entry technique on smartwatches.Moreover,on physical key-boards and smartphone soft keyboards,our algorithm shows significantly higher accuracy than mainstream commercial products when correcting insertion,omission,substitution and transposition errors,which are major kinds of typing errors in various input interfaces.(3)Proposing the hand movement model for ten-finger touch typing:We model the finger kinematics during single-finger tapping procedure,the 3D Gaussian parame-ters of touch point distribution,and especially,the intention revealed by the correlation among fingers.These results significantly increase the prior knowledge in the Bayesian method.Our technique increases the input accuracy of ten-finger touch typing on an in-air imaginary keyboard to nearly 100%,which is a technical breakthrough on the interaction concept of air typing,which was first proposed decades before.(4)Proposing the calculation of word clarity for optimizing the task phrase sets:We quantify the correlation between word clarity and text entry evaluation results(input speed and accuracy).Based on Pareto optimization,our optimized phrase sets show higher score in terms of bigram frequency,word clarity and memorability,which are all key metrics for task phrase sets.Our optimized phrase set with different sizes can serve as public datasets for corresponding text entry studies,which can ensure the external validity of the measured results. |
PDF Full Text Request