Distribution of visual attention revealed by eye movements and neural networks

Abstract

Abstract The present study investigates the relationship between the distribution of visual attention and the information content of selected points of interest in a complex problem-solving situation by eye tracking technology and neural networks. In a first experiment subjects were asked to direct a computer simulated factory. All information necessary for problem management was presented on a projection screen in front of the subject. Four out of twenty information items could be influenced. While the subject was deciding on the economic measures, eye movements were recorded. The results indicated that successful subjects use a more selective information gathering strategy than unsuccessful subjects. In a second experiment the relevance of each item for making decisions was assessed. For this purpose neural networks were trained by error back-propagation. Information, presented in the problem-solving situation before, now provided the input values. The decisions of the subjects determined the four output units. After training the networks, an index for every input unit was formed. These indices, which stand for the contribution of information items on decision making, were correlated with relative fixation frequencies and fixation duration of information items. High correlation coefficients were found for the relationship between the index of items and fixation frequency. The results suggest that the amount of attention, as measured by the number of gaze frequencies, reflects the relative importance of information items. It was also found that problem-solvers used only simple logical rules when deciding on the economic measures.