In this paper, we tackle the problem of daily activities recognition in a multi-resident e-health smart-home using a semi-supervised learning approach based on neural networks. We aim to optimize the recognition task in order to efficiently model the interaction between inhabitants who generally need assistance. Our hierarchical multi-label classification (HMC) approach provides reasoning based on real-world scenarios and a hierarchical representation of the smart space. The performance results prove the efficiency of our proposed model compared with a basic classification task of activities. Mainly, HMC highly improves the classification of interactive activities and increases the overall classification accuracy approximately from 0.627 to 0.831.