Gaze-Assisted User Intention Prediction for Initial Delay Reduction in Web Video Access
Despite the remarkable improvement of hardware and network technology, the inevitable delay from a user’s command action to a system response is still one of the most crucial influence factors in user experiences (UXs). Especially for a web video service, an initial delay from click action to video start has significant influences on the quality of experience (QoE). The initial delay of a system can be minimized by preparing execution based on predicted user’s intention prior to actual command action. The introduction of the sequential and concurrent flow of resources in human cognition and behavior can significantly improve the accuracy and preparation time for intention prediction. This paper introduces a threaded interaction model and applies it to user intention prediction for initial delay reduction in web video access. The proposed technique consists of a candidate selection module, a decision module and a preparation module that prefetches and preloads the web video data before a user’s click action. The candidate selection module selects candidates in the web page using proximity calculation around a cursor. Meanwhile, the decision module computes the possibility of actual click action based on the cursor-gaze relationship. The preparation activates the prefetching for the selected candidates when the click possibility exceeds a certain limit in the decision module. Experimental results show a 92% hit-ratio, 0.5-s initial delay on average and 1.5-s worst initial delay, which is much less than a user’s tolerable limit in web video access, demonstrating significant improvement of accuracy and advance time in intention prediction by introducing the proposed threaded interaction model.
Data Preloading Technique using Intention Prediction
Various smart devices provide fast response time and ubiquitous web-environment to users for better user experiences (UXs). However, high device performance that users perceive is not always promised because there should be limited network bandwidth, and computation capabilities. When the network and computation capabilities are overloaded, users experience buffering and loading time to accomplish a certain task. We, therefore, propose data preloading technique, which predicts user intention and preloads the web and local application data to provide better device performance in spite of poor network conditions and outdated hardware. We also design intention cognitive model to predict user intention precisely. Four user intention prediction algorithms, which are applicable to various conventional input methods, are described and compared each performance in both user’s and device’s aspects.
Seungyup Lee 