Personality Traits Inference in the Hybrid Foraging Search Task

Techniques to predict participants' personality traits in real-time are not yet developed or well-studied. The objective of the current study was to explore the use of gaze and behavioral metrics and machine learning techniques in a hybrid foraging search task to infer an individual's personality traits to enable personalized interaction. We recruited and collected data from 40 university student participants in a hybrid foraging search task experiment. Specifically, the metrics were extracted from different time window sizes (5s, 10s, 15s, and 20s), which referred to the length of time before the participant stopped searching the current screen. Hierarchical clustering analysis was performed on the personality traits scores to group the participants into three groups, namely neuroticism (47.50%), conscientiousness (25.00%), and agreeableness (27.50%). Machine learning models were trained using the eye-gaze and behavioral metrics as inputs and personality trait groups as labels using well-known algorithms (including random forest (RF), support vector machine (SVM), k- nearest neighbor (kNN), and artificial neural network (ANN)). The results from the machine learning modeling showed that the prediction accuracy increased as the window size increased in general. The highest prediction accuracy (83%) was achieved with the kNN algorithm with a 15s time window. Combining eye-gaze and behavioral metrics as input features usually resulted in a better-performing model compared to using eye-gaze metrics alone (up to 10% improvement in accuracy). The current results can be to implement this approach in a brief game-like activity to infer a user's personality traits to enable subsequent intelligent user interface adaptations.