Research on the Application and Optimization of the Multi-Head Self-Attention Mechanism in xDeepFM Personalized Exercise Resource Recommendation

At present, the personalized recommendation system has emerged as a crucial technology to address the issue of cognitive overload and disorientation in the process of online learning. This paper proposes a new fusion structure—Extreme Deep Factorization Machine (xDeepFM) with Multi-Head Self-Attention Mechanism—to capture the relationship between the features of students' learning behavior and the features of exercise resources. By introducing a multi-head self-attention mechanism in front of the deep neural network (DNN) layer, the model was enhanced over xDeepFM. It was used to fully explore the relationship between the various scores of learners in the learning process and the features of exercises in order to offer exercise resources for learners. An actual dataset from an online learning platform was used to test the model that is presented in this research. The model was compared with different recommendation algorithms. The experimental results demonstrated that the proposed model performs much better than other recommendation algorithms in terms of each evaluation indicator.