To address the issue of delay in multi-campus sharing of preschool education resources in colleges and universities, uneven load of nodes, and insufficient precision of personalized recommendation in the smart education cloud platform, this work constructs an integrated system of distributed sharing and personalized recommendation.
Leveraging the national digital education strategy, this initiative focuses on the critical role of preschool education resources (courseware, case studies, etc.) in teacher training within higher education institutions, addressing limitations of traditional management models. The distributed sharing module employs an enhanced distributed hash table (DHT) indexing mechanism, combining dynamic weighted allocation strategies with a lightweight Paxos-Raft Consensus Algorithm to optimize resource scheduling and data synchronization. The personalized recommendation module proposes an attention-based distributed hybrid recommendation (ADHR) model. It integrates semantic modeling from bidirectional encoder representations from transformers with behavioral temporal capture from bidirectional long short-term memory networks, tracking dynamic user demands. It is validated using a sample of 15,000 resources and 1,300 users from three universities within a 10-node cloud environment, supplemented with comparisons against mainstream temporal recommendation models (SASRec and SimSRec) and a methodology explanation for user surveys.
Compared to the traditional Hadoop distributed file system (HDFS), the improved DHT architecture achieves a 34.7% reduction in average response time, a node load balancing coefficient of 0.93, a parallel efficiency of 1.46, and a throughput decline of no more than 10% under abnormal scenarios. Compared to traditional collaborative filtering (CF) algorithms, the ADHR model achieves an 18.9% improvement in accuracy, a 21.2% increase in recall, and reduces root mean square error (RMSE) to 0.191.
This model maintains high performance across scenarios, including cold starts and data sparsity. Overall satisfaction among teachers and students exceeded 90%. Under a scenario with 1,000 concurrent users, the system response time is only 397 milliseconds, with an error rate below 1%. This system effectively enhances resource sharing efficiency and recommendation accuracy, enriches resource management theories in the field of smart education and provides technical support for preschool teacher training. To improve the rigor of experimental evaluation, this study supplements comparisons with mainstream temporal recommendation models, clarifies the positioning of distributed architecture comparisons and details user survey methodologies to enhance result credibility.
