Multi-task Online Course Recommendation Method Based on FDMA

In current online learning platforms, there is limited interaction between learners and courses, leading to issues of data sparsity and noise that significantly impact the recommendation performance of online courses. To address this problem, this paper proposes a feature decomposition multi-task online course recommendation model that integrates the multi-head self-attention mechanism and autoencoder (FDMA). This model adopts a feature decomposition-based multi-task architecture, specifically dividing the recommendation task into two sub-tasks: learner browsing courses and learner effectively learning courses. By optimizing both tasks, the model captures user interests and preferences while satisfying user real needs and learning experiences. The model utilizes a multi-head self-attention mechanism for automatic feature interaction and extraction, and an autoencoder for feature dimensionality reduction and noise reduction, greatly enhancing the recommendation performance of the model. Experimental results demonstrate that FDMA outperforms baselines, achieving improvements of 0.27% in ACC, 0.04% in AUC, and a 35.66% reduction in LogLoss, validating its effectiveness and superiority.