Sentiment analysis of ethnic artworks using deep learning

Abstract

Abstract Deep learning models, characterized by their complex network layers, have demonstrated significant utility in the recognition and classification of ethnic artworks, particularly excelling in emotion recognition within such works. This study delves into the convolutional neural network (CNN) model underpinned by deep learning algorithms. The architecture of the model—encompassing the convolutional layer, pooling layer, fully connected layer, and activation layer—is meticulously constructed to extract emotional features from ethnic artworks. Subsequently, an activation function is employed to visualize these features, followed by the integration of emotional knowledge to optimize the network structure. The training of the model utilizes both the loss function and optimization function to enhance performance. The efficacy of the sentiment analysis is assessed through simulation experiments and practical applications, highlighting the model’s superior capability in identifying negative emotions such as Fear and Anger, achieving accuracies that are 0.101 and 0.137 higher, respectively, than those of the benchmark model. Additionally, a detailed analysis of the emotion scores of ethnic artworks reveals intriguing findings. Notably, Jiye Li’s “Four Seasons of Essays” registers the highest positive mean emotional score of 4.1026, with an overall mean of 3.5392, indicating a predominantly positive emotional response. Across the works of nine subjects analyzed, the aggregate mean emotional score stands at 1.0123, suggesting a generally positive emotional tone. This research underscores the nuanced capability of CNNs in the domain of emotional recognition in ethnic art, offering insights into both methodological advancements and interpretive analyses.