设备端基于深度学习的智能家居服务推荐框架

随着智能家居的普及，用户期望通过自然语言指令实现智能设备的控制，并希望获得个性化的智能家居服务。然而，现有的挑战包括智能设备的互操作性和对用户环境的全面理解。针对上述问题，提出一个支持设备端用户智能家居服务推荐个性化的框架。首先，构建智能家居的运行时知识图谱，用于反映特定智能家居中的上下文信息，并生成用例场景语句；其次，利用预先收集的通用场景下，用户的自然语言指令和对应的用例场景语句训练出通用推荐模型；最后，用户在设备端以自然语言管理智能家居设备和服务，并通过反馈微调通用模型的权重得到个人模型。在基本指令集、复述集、场景指令集三个数据集上的实验表明，用户的个人模型相比于词嵌入方法的准确率提升了6.5%~30%，与Sentence-BERT模型相比准确率提升了2.4%~25%，验证了设备端基于深度学习的智能家居服务框架具有较高的服务推荐准确率，能够有效地管理智能家居设备和服务。

On-device deep learning for smart home service recommendation framework

As smart homes become more prevalent, users expect to control smart devices through natural language commands and desire personalized smart home services. However, existing challenges include the interoperability of smart devices and a comprehensive understanding of the user environment. To address these issues, this paper proposed a framework supporting personalized smart home service recommendations for device-end users. Firstly, it constructed a runtime knowledge graph to reflect contextual information in specific smart homes and generated scenario-based sentences. Secondly, it trained a general recommendation model using pre-collected natural language instructions and corresponding scenario-based sentence representations from users in common scenarios. Finally, users interacted with smart home devices and services through natural language on the device end while fine-tuning the weights of the general model through feedback to obtain a personal model. Experimental results on three datasets-basic instruction set, paraphrase set, and scenario instruction set show that the personal model achieves an accuracy improvement of 6.5% to 30% compared to word embedding methods and 2.4% to 25% compared to the Sentence-BERT model, which validates that the device-end deep learning-based smart home service framework has a high service recommendation accuracy and effectively manages smart home devices and services.