Predicting Word Fixations in Text with a CRF Model for Capturing General Reading Strategies among Readers

摘要

Human gaze behavior while reading text reflects a variety of strategies for precise and efficient reading. Nevertheless, the possibility of extracting and importing these strategies from gaze data into natural language processing technologies has not been explored to any extent. In this research, as a first step in this investigation, we examine the possibility of extracting reading strategies through the observation of word-based fixation behavioT. Using existing gaze data, we train conditional random field models to predict whether each word is fixated by subjects. The experimental results show that, using both lexical and screen position cues, the model has a prediction accuracy of between 73% and 84% for each subject. Moreover, when focusing on the distribution of fixation/skip behavior of subjects on each word, the total similarity between the predicted and observed distributions is 0.9462， which strongly supports the possibility of capturing general reading strategies from gaze data.