Prediction of cognitive bugs in mobile computing environments by using a neural network-based path processing model

Abstract

People tend to use mobile applications in their daily life while performing other activities at the same time. In such a case, we define a cognitive bug as the high-intensity interference due to the simultaneous use of cognitive resources by multiple activities that causes task error or interruption, preventing users from fulfilling their goals. Because of the negative consequences of multitasking in mobile environments are well established, the design and development of applications that humans interact with should consider the minimization of cognitive bugs to avoid usage failures. With that purpose, we present a theoretical model for predicting cognitive bugs in mobile applications. Studies that have addressed problems related to prediction or estimation of human cognitive factors have proposed empirical methods for estimating cognitive load, as a metric of observed cognitive effort, collecting the various type of data, using a diverse range of sensors, and developing machine learning models of estimation. However, in mobile application environments, there are several constraints regarding data collection. Thus, in a high mobility environment, to collect good quality data from people that can be correlated with cognitive factors, such as physiological signals, is technically challenging and from people's perspective is uncomfortable and inconvenient to wear different type of sensors. For that reason, we developed a model for simulating cognitive processes and predicting, from a theoretical perspective, cognitive resources usage and cognitive bug occurrence. We developed an ontology model, cognitive task representation model, and simulation of cognitive resource utilization using a parallel neural network architecture. We enriched the neural network architecture with external knowledge, from the ontology model and from a cognitive control mechanism that establishes theoretical rules based on cognitive psychology models. To validate our model we performed a user study in which we tested people while using mobile applications and performing physical activities at the same time. Through a statistical comparison against empirical estimation methods, results show 85% of correlation on average with our model. Furthermore, we create metrics to define cognitive bugs and provide information on which tasks can be performed in parallel avoiding cognitive bug occurrence. In addition, task classification results of our model indicate that assuming a sequential structure in the neural network the average accuracy is 90%.