Exploration of art portrait classification and intelligent education model based on unsupervised emotion recognition in online education

Artistic portraiture holds a significant position in the art world, with its stories of artistic personalities carrying significant educational value and holding offers much scope for research. Amidst the contemporary artistic education milieu, pedagogical practices have embraced an amalgam of online and offline intelligent instruction modes, thus broadening the art students’ access to artistic portraiture. However, the abundance of digital resources has created a concomitant conundrum regarding the classification and selection of artistic portraiture, emerging as a pressing issue. Emotion recognition, a mechanism that enables natural human-computer interaction, assumes critical importance in artistic portraiture classification, as character emotional expressions serve as a pivotal basis for the said classification. Nevertheless, the extensive costs incurred during the collection and annotation of emotional data related to artistic portraiture constitute a major bottleneck that inhibits the development of emotion recognition research. This paper endeavors to explore the application of knowledge transfer between cross-domains or cross-tasks to enhance the efficacy of emotion recognition in scenarios involving limited or no labeling. To achieve efficient classification of portraiture, the paper employs a blend of domain adaptive and transfer learning techniques to map features onto a shared space between domains. Finally, the effectiveness of this method is verified through unsupervised sentiment classification and labeling of artistic portraiture based on the collected dataset, which resulted in an improvement in performance, achieving a classification accuracy of 46.7% compared to the most relevant domain adaptive sentiment analysis methods.