Exécution de notre outil de synthèse sur un nouvel article
Vérifiez l’article ci-dessous, puis lancez sa synthèse.
Data modeling of English classroom interaction based on deep learning to support evidence-based education decision-making
Yanbei Wu (College of Modern Science and Technology, China Jiliang University, China Jiliang University), Caihui Guo (Guangzhou Nanyang Polytechnic College)
Résumé : Introduction: classroom interaction is very important for learning, especially in language classes where communication, feedback, and engagement are key to skill development. However traditional methods of watching and evaluating students in the classroom are typically subjective, time-consuming, and not easy to scale, which makes it challenging to make decisions about education based on data. Methods: this research suggests a data modeling approach based on deep learning (DMA-DL) to examine how people interact in English classrooms to help make decisions about education that are based on data and evidence. The suggested structure combines together several types of classroom data, such as audio recordings, video feeds, and written transcripts of conversations between teachers and students. Advanced deep learning architectures are used to automatically find semantic, temporal, and behavioral aspects that show how people interact, like turn-taking, questioning methods, response delay, and learner participation. Recurrent and attention-based models are employed to represent temporal dependencies, whereas convolutional neural networks facilitate comprehensive evaluation of auditory and visual signals.Results: the approach additionally utilizes representation learning to produce interaction indicators that are easy to comprehend and compatible with educational theories. The collected interaction elements are combined into a structured data model that allows it evaluates how well the teaching methods and student engagement are working. Experimental study performed on authentic classroom datasets illustrates that the suggested methodology attains significant accuracy in recognizing interaction patterns and proficiently differentiates between various teaching styles and levels of engagement. DMA-DL attains an overall interaction classification accuracy of 94.6%, outperforming DBDM (88.3%), SLR (86.9%), and EBNE (90.5%). In addition, the model records precision of 93.8%, recall of 94.1%, and an F1-score of 94.0%, indicating consistent predictive reliability across multiple interaction categories.Conclusions: the analytics provide teachers and administrators useful information that helps them make decisions about things like improving the curriculum, designing lessons, and professional development.