Attention-LGBM-BiLSTM: An Attention-Based Ensemble Method for Knowledge Tracing
Si Shi, Wuman Luo, Rita T. Tse, Giovanni Pau · 2022
Knowledge tracing plays a vital role in measuring students’ learning behaviors. In this paper, we propose a novel ensemble model: Attention-LGBM-BiLSTM for knowledge tracing. We utilize the attention mechanism combined with LGBM (Light Gradient Boosting Machine) to obtain a feature of the most importance. Combined with the first-round outputs of LGBM, it is imported into BiLSTM (Bidirectional Long Short-Term Memory), thus obtaining the final classification results. We implement and evaluate the model based on the largest open-source dataset, EdNet, in education area. The results show that the accuracy, AUC, and Fl-score of the model are higher than its baselines. An ablation test is also conducted to prove its effectiveness.