Model Fusion of LightGBM and SAKT for Knowledge Tracking
Xin Zhou, Liming Zhang, Fanqi Meng · 2021
Knowledge tracking is to model the student’s learning interaction records so that we can evaluate the student’s learning state with relative accuracy and can adjust the students’ learning schedule appropriately or formulate a more reasonable learning plan. However, this task is very challenging. The challenges are mainly in two aspects. On the one hand, each individual is different, and on the other hand, handling data is complex. Currently, there are two main ideas for knowledge tracking, namely machine learning or deep learning. For such ”knowledge tracking” problems, our idea is to combine traditional machine learning solutions and neural network sequence models to obtain better results in such problems. We proposed the LGBM+SAKT model fusion method, which organically combines machine learning and deep learning, which not only makes full use of as many data features as possible, but also extracts and represents data features at a deep level without losing readability. We complete this task in a knowledge tracking competition on Kaggle, scoring at least 1.8% over every single model at the end.