Abstract

This article is devoted to solving the problem of classifying a patient’s emotions in nonverbal communication based on machine learning. The article reflects the current state of the problem in the field of classification of human emotions in nonverbal communication and gives the formulation of the problem and the results of experimental studies for finding the most high-quality model that allows for the classification of emotions. The resulting model has an architecture that includes two recurrent subnets with an attention mechanism, the outputs of which are combined and fed to a fully connected classification layer. As a result, the obtained model on the validation dataset has accuracy of 0.77. The model was developed in Python using the TensorFlow and Keras frameworks. To extract the image of the face, the BlazeFace model from the MediaPipe framework was used.

中文翻译：

---

基于机器学习的非语言交流中患者情绪分类

摘要

本文致力于解决基于机器学习的非语言交流中患者情绪的分类问题。本文反映了非语言交流中人类情感分类领域问题的现状，并给出了问题的表述和实验研究的结果，以寻找允许情感分类的最高质量模型。所得模型的架构包括两个带有注意机制的循环子网，其输出被组合并馈送到完全连接的分类层。结果，在验证数据集上获得的模型的准确度为 0.77。该模型是使用 TensorFlow 和 Keras 框架用 Python 开发的。为了提取人脸图像，