<p>Mobile Instant Messaging Online Health Communities (MIM OHCs) provide real-time healthcare guidance and support, particularly for major or chronic illness patients. However, the diverse discussions among users in MIM OHCs challenge new members' engagement in dialogues, while semi-structured messages pose hurdles for researchers in extracting insights. Consequently, it is crucial to uncover the underlying dialogues in MIM OHCs. Several key limitations are worth noting. Firstly, complex deep learning models tend to overfit or underfit when trained on limited labeled data. And the scarcity of suitable text data for domain adaptation hampers optimal recognition. Secondly, existing research often emphasizes text semantics while ignoring social support information. To this end, we introduce the Social-support-aware Dialogue Recognition in Mobile Instant Messaging Online Health Communities model (SDRMO). The SDRMO employs a Multi-BiLSTM for social support information extraction and a domain-adapted BERT model trained on well-structured OHC data for semantic information extraction. The SDRMO adopts a semi-supervised learning approach with a pseudo-labeling method to enhance the model's robustness and generalization. A fully connected neural network facilitates the fusion of both social support and semantic information. Comprehensive experiments and ablation studies confirm SDRMO's superior performance in dialogue recognition within MIM OHCs.</p>