Çizim temelli arayüzlerde tanıma hatalarının göz hareketleri kullanılarak belirlenmesi

Sketch based intelligent interfaces are gaining popularity as pen based hardware becomes more widespread. These interfaces make use of sketch recognition technology to facilitate natural and efficient interaction. Nevertheless all sketch recognition systems suffer from misrecognitions, which inflicts a correction cost on to the user. Every time a symbol gets misrecognized, the user explicitly or implicitly signals his intention to correct the error, and does so by either redrawing the symbol or selecting it from a list of alternatives. We propose a system for alleviating the cost of this two-step process by detecting users' intention to fix misrecognitions based on their eye gaze activity. In particular, we show that users' natural reaction to misrecognitions manifests itself in the form of characteristic eye gaze movement patterns. Furthermore, these patterns can be used to read users' intention to fix errors before they initiate such action. We have three main contributions. First, we present a carefully constructed Wizard of Oz setup for recording eye gaze patterns under two sketch-based interaction conditions. Then, we present a set of gaze-based features, which were designed to capture qualitative characteristics of users' eye gaze behavior. Finally, we present a framework for recognizing users' intention to fix errors, which achieves an 86% prediction accuracy. We support our findings through detailed experiments and statistical analyses, which provide further insight into how much can be inferred from eye gaze patterns that naturally emerge during pen-based interaction.