Predicting Multiple Reading Tasks Using Eye Movement Measures

Understanding user engagement and behavior during reading activities is important for multimedia system design as it can provide valuable insights into developing more effective interactive systems. This paper proposes a novel approach to predicting multiple reading tasks based on eye movement measures. We conducted experimental research involving participant trials across various reading tasks, collecting time-series eye movement data. Our method employs Convolutional Neural Networks (CNNs) with Bidirectional Long Short-Term Memory (BiLSTM) to identify patterns associated with four reading tasks: skimming, reading, scanning, and proofreading. Importantly, this method doesn't require experts to label local windows manually and requires minimal data pre-processing, as the model autonomously extracts features. The proposed model demonstrates promising performance, achieving a multiclass classification accuracy of $\mathbf{7 0. 0 4 \%}$. The experimental results indicate that the proposed model surpasses the performance of the other methods examined in the study. Furthermore, our proposed method holds the potential for real-time reading task classification, which can deepen our understanding of user behavior and interaction patterns. By enabling accurate prediction of reading tasks, our approach facilitates the development of enhanced multimedia browsing and search systems tailored to user preferences and goals.