Multimodal Dataset for Personality Prediction

ABSTRACT: Personality as a stable pattern of behavior, thoughts, and emotions, affects individuals' daily life. However, existing datasets rarely focus on personality prediction from multi-modal signals, hindering the progress in personality computing from the neurophysiological perspective. Given that existing research has also reveal the intrinsic interconnection between personality and emotion, we present a multi-modal dataset with four kinds of signals recorded during participants' viewing of different emotion videos. To measure individual personality traits, the BFI-44 questionnaire was used as the personality indicator. Then, an experiment with 63 participants was conducted, in which the data of EEG, GSR, PPG, and face video were collected when they watched the movie video clips. We also recorded the subjective PANAS, VAD and emotion rating. Totally, the dataset consists of multi-modal signal data and self-assessment data. We performed technical validation based on the participants' subjective assessment data and the collected signal data to validate the effectiveness of emotion induction and the quality of the collected signal data. Moreover, we conducted extensive experiments for personality prediction using different modality combinations and four classifiers (kNN, SVM, RF, and MLP). When using RF classifier and three psychological modality features, the 5-fold cross-validation average accuracy for the five personality trait binary classification tasks across seven emotions can reach approximately 90%, demonstrating the potential of objective personality prediction.