Cross-platform adult learning behaviour profiling based on multimodal data fusion

In the current era of rapidly developing digital education, single-modal data is insufficient to accurately characterise student learning behaviour.Cross-platform data contains rich information but faces challenges such as data silos and modal heterogeneity.To address this, this paper first collects multi-modal learning behaviour data from students, employs deep learning models to extract text and visual features, and uses attention to fuse multi-modal data.The multimodal fused data is then carefully segmented and distributed to different platform stakeholders.These stakeholders interact and aggregate information through a federated learning framework to obtain more accurate representations of learning behaviour features.Based on this, a Softmax classifier is used to precisely classify cross-platform student learning behaviour profiles.Experimental results show that the proposed method achieves an MAE of only 0.0631, significantly improving the classification accuracy of learning behaviour profiles.