Gutiérrez-Martín Laura, López-Ongil Celia, Lanza-Gutiérrez Jose M, Miranda Calero Jose A
Departamento de Tecnología Electrónica, Universidad Carlos III de Madrid, Avenida de la Universidad, 30, 28911 Leganés, Spain.
Instituto de Estudios de Género, Universidad Carlos III de Madrid, Calle Madrid, 126, 28903 Getafe, Spain.
Sensors (Basel). 2024 Dec 19;24(24):8110. doi: 10.3390/s24248110.
Emotion recognition through artificial intelligence and smart sensing of physical and physiological signals (affective computing) is achieving very interesting results in terms of accuracy, inference times, and user-independent models. In this sense, there are applications related to the safety and well-being of people (sexual assaults, gender-based violence, children and elderly abuse, mental health, etc.) that require even more improvements. Emotion detection should be done with fast, discrete, and non-luxurious systems working in real time and real life (wearable devices, wireless communications, battery-powered). Furthermore, emotional reactions to violence are not equal in all people. Then, large general models cannot be applied to a multi-user system for people protection, and health and social workers and law enforcement agents would welcome customized and lightweight AI models. These semi-personalized models will be applicable to clusters of subjects sharing similarities in their emotional reactions to external stimuli. This customization requires several steps: creating clusters of subjects with similar behaviors, creating AI models for every cluster, continually updating these models with new data, and enrolling new subjects in clusters when required. An initial approach for clustering labeled data compiled (physiological data, together with emotional labels) is presented in this work, as well as the method to ensure the enrollment of new users with unlabeled data once the AI models are generated. The idea is that this complete methodology can be exportable to any other expert systems where unlabeled data are added during in-field operation and different profiles exist in terms of data. Experimental results demonstrate an improvement of 5% in accuracy and 4% in F1 score with respect to our baseline general model, along with a 32% to 58% reduction in variability, respectively.
通过人工智能以及对身体和生理信号的智能感知来进行情感识别（情感计算），在准确性、推理时间和独立于用户的模型方面正取得非常有趣的成果。从这个意义上说，与人们的安全和福祉相关的应用（性侵犯、基于性别的暴力、虐待儿童和老人、心理健康等）还需要进一步改进。情感检测应该通过快速、离散且不昂贵的系统在现实生活中实时运行（可穿戴设备、无线通信、电池供电）来完成。此外，对暴力的情感反应在所有人中并不相同。因此，大型通用模型不能应用于用于人员保护的多用户系统，健康和社会工作者以及执法人员会欢迎定制化且轻量级的人工智能模型。这些半个性化模型将适用于在对外部刺激的情感反应上具有相似性的主体集群。这种定制需要几个步骤：创建具有相似行为的主体集群，为每个集群创建人工智能模型，用新数据持续更新这些模型，并在需要时将新主体纳入集群。本文提出了一种对编译的标记数据（生理数据以及情感标签）进行聚类的初始方法，以及在生成人工智能模型后确保将未标记数据的新用户纳入的方法。其理念是，这种完整的方法可以推广到任何其他在现场操作期间添加未标记数据且存在不同数据概况的专家系统。实验结果表明，相对于我们的基线通用模型，准确率提高了5%，F1分数提高了4%，同时变异性分别降低了32%至58%。
