The aggravation of population aging makes the demand for elderly care expanding. In this paper, we propose an integrated care model based on deep learning to build an intelligent service robot system for elder care organizations by integrating sentiment analysis and knowledge reasoning techniques. The model is driven by the dynamic needs in long-term care scenarios, and two modules are innovatively designed. In the sentiment analysis module, multimodal sensors (facial expression, audio state, textual content) and graph attention networks are integrated, and global contextual information is modeled on these features to identify long-distance emotional dependencies of the elderly. In the knowledge inference module, graph representation learning is combined with knowledge graph temporal inference to construct an inference model to speculate the care needs of the elderly. The experiment shows that after the system performs long-term service, the depression condition of the elderly is significantly improved, and the nursing care safety risk perception shows a significant difference from that before the system is used (P<0.001). The integrated care model studied in this paper provides a practical technical solution to the problem of aging care resource shortage.
