Predict the writer's trait emotional intelligence from reproduced calligraphy

Abstract

Trait emotional intelligence (EI) describes an individual's ability to control their emotions. In Chinese calligraphy, there is a saying that "the character reflects the person." This raises a hypothesis: is it possible to predict a writer's trait EI from their calligraphy reproductions? To test this hypothesis, we propose a predictive method that integrates deep learning with aesthetic features of calligraphy. First, a hard pen calligraphy reproduction dataset was constructed, consisting of 48,826 reproduced characters from 191 participants, with corresponding trait EI scores and reproduction skill score ratings. A Siamese neural network was then used to extract deep feature differences between the reproduction characters and the reference characters, which were further combined with handcrafted features for regression-based predictions. Experimental results show that, using Mean Absolute Error (MAE), Mean Squared Error (MSE) and Pearson Correlation Coefficient (PCC) as evaluation metrics, this method's ability to predict the writer's trait EI from calligraphy reproductions (MAE: 0.463, MSE: 0.462, PCC: 0.730) significantly outperforms human evaluative abilities (MAE: 1.006, MSE: 1.740, PCC: 0.145), confirming that calligraphy reproductions indeed contain latent information about the writer's trait EI.

Keywords: Calligraphy psychology; Computational aesthetics; Computer assisted assessment; Psychological projection experiment; Siamese neural network; Trait EI prediction.

Fig. 1

Do the differences in the reproductions by three writers reveal their trait emotional intelligence? (Red: reference character, Blue: different areas of the reference and reproduced character, Black: overlap areas of the reference and reproduced character).

Fig. 2

Research design.

Fig. 3

The pre-selected 55 high-frequency Chinese characters have different stroke numbers, covering different levels of complexity and 11 common structures.

Fig. 4

The preprocessing of image data in data set.

Fig. 5

The violin diagram for score distribution of each dimension data after conversion to z-score.

Fig. 6

Prediction model architecture diagram.