Complex network of eye movements during rapid automatized naming

Abstract

Introduction: Although the method of visualizing eye-tracking data as a time-series might enhance performance in the understanding of gaze behavior, it has not yet been thoroughly examined in the context of rapid automated naming (RAN).

Methods: This study attempted, for the first time, to measure gaze behavior during RAN from the perspective of network-domain, which constructed a complex network [referred to as gaze-time-series-based complex network (GCN)] from gaze time-series. Hence, without designating regions of interest, the features of gaze behavior during RAN were extracted by computing topological parameters of GCN. A sample of 98 children (52 males, aged 11.50 ± 0.28 years) was studied. Nine topological parameters (i.e., average degree, network diameter, characteristic path length, clustering coefficient, global efficiency, assortativity coefficient, modularity, community number, and small-worldness) were computed.

Results: Findings showed that GCN in each RAN task was assortative and possessed "small-world" and community architecture. Additionally, observations regarding the influence of RAN task types included that: (i) five topological parameters (i.e., average degree, clustering coefficient, assortativity coefficient, modularity, and community number) could reflect the difference between tasks N-num (i.e., naming of numbers) and N-cha (i.e., naming of Chinese characters); (ii) there was only one topological parameter (i.e., network diameter) which could reflect the difference between tasks N-obj (i.e., naming of objects) and N-col (i.e., naming of colors); and (iii) when compared to GCN in alphanumeric RAN, GCN in non-alphanumeric RAN may have higher average degree, global efficiency, and small-worldness, but lower network diameter, characteristic path length, clustering coefficient, and modularity. Findings also illustrated that most of these topological parameters were largely independent of traditional eye-movement metrics.

Discussion: This article revealed the architecture and topological parameters of GCN as well as the influence of task types on them, and thus brought some new insights into the understanding of RAN from the perspective of complex network.

Keywords: complex network; developmental dyslexia; eye tracking; rapid automatized naming; time series.

Figure 1

The Chinese RAN paradigm presented a 5 × 10 matrix of stimuli (e.g., numbers, Chinese characters, colors, or objects) in different tasks: (A) Task N-num (i.e., naming of numbers); (B) Task N-cha (i.e., naming of Chinese characters); (C) Task N-col (i.e., naming of colors); and (D) Task N-obj (i.e., naming of objects). RAN, rapid automatized naming.

Figure 2

Method, involving three steps, to construct the gaze-time-series complex network (GCN) in an illustrating example of a child during a RAN task. (A) Heatmap of gazes in the illustrating example; (B) Raw spatial coordinate of eye gazes in the illustrating example, where two adjacent gazes were linked by a straight line; (C) The network connectivity of GCN in the illustrating example, where dots represented nodes and black lines represented connections between nodes; (D) Topological properties analysis and their values in the illustrating example. RAN, Rapid automatized naming.

Figure 3

Influence of task types on different topological parameters: (A) average degree; (B) network diameter; (C) characteristic path length; (D) clustering coefficient; (E) global efficiency; (F) assortativity coefficient; (G) modularity; (H) community number; and (I) small-worldness. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001. RAN, rapid automatized naming; N-num, naming of numbers; N-cha, naming of Chinese characters; N-col, naming of colors; N-obj, naming of objects.