Gaze Convergence Based Collaborative Performance Prediction in a 3-Member Joint Activity Setting

Teams engaged in desktop-based collaborative problem solving has been increasing in recent years. Multiparty joint activities such as computer-supported collaborative working (CSCW) demands intelligible communication for efficient coordination and cooperation. Although, studies have demonstrated a relationship between members’ gaze and performance, the practicality of application in larger groups and intelligent predictive modelling has shown slow progress. One common constraint observed in literature is that much of the studies used dyadic team setting, and when larger groups are considered studies have resorted to a laborious examination of videos. Further, in the context of performance prediction, regression and binary classification are observed in the literature. This provides a motivation to explore neural network based methods, however, it demands the selection of appropriate team members’ gaze features. This study aims to advance gaze-based collaborative research by overcoming some of the drawbacks. In this light, a desktop-based 3-member collaborative activity is adopted in the study to facilitate our investigations. Linear statistics were explored to observe gaze convergence and observed a significant correlation with task performance. The measures were used to model a multivariate regression equation and a neural network, and the mean absolute errors of both models were compared. The comparison revealed promising results suggesting the employment of team members’ gaze behaviour for objective assessment and prediction of collaborative performance.