Visual Attention-Guided Weighted Naïve Bayes for Behavior Intention Inference

Intention inference based on eye gaze is an important issue in human-computer interaction. To address the issue of neglecting feature weight differences in behavior intention inference algorithms using visual attention objects as input features, a visual attention-guided weighted naïve Bayes behavior intention inference algorithm is proposed. Firstly, experiments are conducted to explore the eye-movement pattern for different behavioral intentions. Then, a method for quantifying visual attention using spatial-temporal features of eye-movement data is introduced. Finally, a visual attention prior weight matrix is constructed to guide the modified class-specific attribute weighted naïve Bayes model to learn the feature weights of different visual attention objects for achieving behavior intention inference. The experimental results show that the accuracy of the proposed method is 95.31%, which is at least 2.6% higher than the accuracy of the existing methods. At the same time, it can suppress the influence of visual attention noise objects to a certain extent, improving the robustness of the algorithm.