Modeling of Drama Performance Intelligent Evaluation Driven by Multimodal Data

The purpose of this study is to explore a data-driven intelligent evaluation method for drama performances, and to improve the evaluation quality of drama performances. Our research work is mainly to establish the temporal relationship between motion and musical features in dramatic performances and to construct a multimodal evaluation dataset (PEMD, Performance Evaluation Multimodal Dataset) for drama performances based on computer vision methods and deep learning technologies. Then the evaluation of drama performance is achieved by detecting and evaluating the match degree (DMD, Dramatic Match Degree) of the motion and musical features in the drama performance. The main works includes: (1) A data-driven intelligent evaluation framework for drama performance is proposed, which defines and describes the collection method, classification and feature extraction of drama performance evaluation data; (2) A sliding window computing unit based on Dramatic Stylization Annotation is proposed. As the core computing module of the drama performance evaluation architecture, it establishes the corresponding relationship between performance motion and music based on temporal features, and constructs Multimodal Evaluation Dataset (PEMD) for the drama performance; (3) Aiming at the temporal features of drama performances, a co-training method is proposed to establish the Theater Performance Evaluation Model (TPEM) and realize intelligent computing methods for drama performance intelligent evaluation. The experimental results show that the average accuracy rate (MAP Mean Average Precision) of the drama performance evaluation model proposed in this paper reaches 62.41%, showing excellent evaluation ability.