As the world’s population ages, the issue of medical care and daily care for the elderly population is becoming more and more critical. While there are many efforts to ensure the physical health of older adults, there is a lack of effective solutions to the psychological problems caused by the lack of companionship. To solve this problem, we built a social robot with an emotion recognition and feedback model. The emotion recognition module achieves composite emotion recognition by integrating voice emotion recognition, semantic emotion recognition and topic emotion recognition. The audio data and corresponding text data are mined using CNN and LSTM algorithms, and the emotional information they contain can be further refined. Meanwhile, the emotion feedback module provides the appropriate emotion feedback based on the recognised emotions. The emotional feedback is achieved by generating the appropriate robot facial expressions. The proposed emotion recognition and feedback model is validated by the corresponding database respectively. In addition, the real-world implementation validation and real scenario application of the model are also discussed in the paper.