Abstract Mind wandering is a state in which an individual’s attention is decoupled from the task at hand. Mind wandering affects performance in many tasks requiring focused attention, including online learning. Previous studies have focused primarily on investigating mind wandering in contexts which are conducive to mind wandering, that is, for highly repetitive and monotonous tasks or during tasks with very low attentional demands. However, mind wandering also occurs during highly engaging and demanding tasks. In this study, we examine whether mouse tracking can be used to predict mind wandering in an engaging task involving classical computer interfaces. Assuming that mouse trajectories towards a particular response on the screen are continuously updated by time-dependent and temporally-dynamic cognitive processes, as a behavioral methodology, mouse tracking can provide unique insights into attentional processes. In our experiment, a total of 272 students completed a mouse-based operation span task, during which their thoughts were probed and their mouse movements recorded. Naive Bayes, Linear Discriminant Analyses, K-Nearest Neighbors, Tree Bag, and Random Forest classifiers were able to predict mind wandering with F1-scores of up to 15% above a random-chance baseline. The results show that hand reach movements can be tracked to detect mind wandering in a user-independent manner in online tasks, thus providing a viable alternative to self-report methods and (neuro)physiological measures. Our finding has relevant implications for a variety of user interfaces which require hand and finger movements for the purposes of human–computer interactions.

中文翻译：

---

流浪老鼠，流浪思想：使用计算机鼠标跟踪来预测思想流浪

摘要 走神是一种个体注意力与手头任务脱钩的状态。走神会影响许多需要集中注意力的任务的表现，包括在线学习。以前的研究主要集中在有利于走神的环境中调查走神，即高度重复和单调的任务或注意力需求非常低的任务。然而，在高度参与和要求高的任务中也会发生走神。在这项研究中，我们研究了鼠标跟踪是否可用于预测涉及经典计算机界面的引人入胜的任务中的思维游移。假设鼠标对屏幕上特定反应的轨迹通过依赖于时间和时间动态的认知过程不断更新，作为一种行为方法，鼠标跟踪可以提供对注意力过程的独特见解。在我们的实验中，共有 272 名学生完成了一项基于鼠标的操作跨度任务，在此任务中，他们的思想被探查并记录了他们的鼠标动作。朴素贝叶斯、线性判别分析、K 最近邻、树袋和随机森林分类器能够以比随机机会基线高出高达 15% 的 F1 分数来预测思维游移。结果表明，可以跟踪手的伸手动作，以独立于用户的方式检测在线任务中的思维游移，从而为自我报告方法和（神经）生理测量提供了一种可行的替代方法。我们的发现对需要手和手指移动以实现人机交互的各种用户界面具有相关意义。