HiEnWrite: A Hindi-English Bilingual Dataset for Big Five Personality Detection

Detecting human personality traits is a critical task across various domains, including healthcare, education, and psychology. Recent advancements in artificial intelligence have greatly enhanced the automatic detection of personality traits using writing styles and handwriting analysis. However, existing datasets and studies predominantly focus on English, limiting the generalizability to multilingual contexts. To bridge this gap, this study presents HiEnWrite, a novel Hindi–English bilingual dataset comprising handwritten samples collected from over 400 authors, annotated with Big Five personality scores. The dataset consists of approximately 800 images, annotated through rigorous Big Five questionnaires. We propose and evaluate an end-to-end convolutional neural network (CNN) tailored to detect five core personality traits. Additionally, extensive experimentation is conducted using transfer learning-based approaches with multiple CNN architectures, achieving a maximum correlation of 0.411 with VGG19. The model performance is thoroughly assessed using metrics like root mean square error (RMSE), training/testing durations, and residual analysis, demonstrating the efficacy of the proposed approach.