Measuring Trust at Zero-acquaintance: A Cross-cultural Study between Malaysians and Hungarians

Trust is critical in communication, especially in a cross-cultural setting. However, there is still not much open source annotated data for trust studies, which may be due to the lack of a suitable annotation platform or a reliable method for measuring trust. This article aims to fill this gap by presenting a pilot study with our annotation platform. Specifically, we asked participants to rate perceived trustworthiness on short emotional videos from the RAVDESS dataset at zero-acquaintance. The results confirm two key findings that should be considered when rating emotional videos: the emotion and the order in which the videos are displayed. After auditing our annotated data, we have also identified two types of bias that might skew our model's predictions: a task bias and a cultural bias. To understand their potential impact when building trustworthiness prediction models, we analyzed the data statistically and developed pilot trustworthiness' regressors models. As for the cross-cultural statistical analysis, Spearman's correlation matrices show that Malaysian and Hungarian cultures attach similar importance to some traits when evaluating perceived trustworthiness, e.g., authenticity and attractiveness. However, they differ in others (such as expressiveness or friendliness). Hungarians place more importance on friendliness, while Malaysians rely more on expressiveness. Regarding the models, the top regressor achieved a 0.62 correlation value between predictions and ground truths. This fact supports the adequacy of our methodology to annotate videos in terms of trustworthiness.