MM-CrowdNet is a multimodal deep learning system that predicts crowd risk levels by combining visual features from Hajj surveillance videos with sentiment scores from Reddit. It uses CNNs, LSTM, and attention to process video frames and adjusts predictions using social media sentiment.