Intelligent Recommendation System for Academic Competitions Based on Multi-Head Attention and Bidirectional LSTM

To address the issues of inaccurate user interest understanding and insufficient recommendation precision in academic competition recommendation systems, this paper proposes an intelligent recommendation system model based on multi-head attention mechanism and Bidirectional Long ShortTerm Memory (BiLSTM). Compared with traditional recommendation algorithms using expert-defined features, our system demonstrates significant advantages: 1) The integration of BiLSTM and multi-head attention mechanism effectively captures multi-dimensional training history information and accurately models user states. 2) The system adapts dynamic strategies such as Elo rating and weak module identification to effectively meet personalized recommendation needs. Experimental results show that the system achieves: $\mathbf{2 4. 0 9 \%}$ improvement in hit rate for user interest prediction tasks compared to traditional methods, 20.10% coverage rate for recommended questions information, and 10.98% increase in users’ problem-solving accuracy after using the system. This system development provides new insights for intelligent recommendation in academic competitions and contributes to advancing personalized recommendation technologies.