AI GEN ASSISTMENT: ANALYZING THE EFFECTIVENESS OF GENERATIVE AI DATA AMPLIFICATION FOR DKT
Keywords: Deep Knowledge Tracing(DKT), Generative AI, Educational Data Mining
Abstract: In edtech, it is one of the important factors to predict the learning level of learners and whether they will solve problems in the future by utilizing a learner track- ing system. However, the scarcity of initial data limits our ability to make accu- rate predictions from scratch and to evaluate the model’s performance effectively. Therefore, our research team utilized generative artificial intelligence technology to amplify virtual log data in large quantities based on small original data to create a virtual test data set and verify whether it is effective as a learning and evalua- tion data set for DKT Model. As a result of the experiment, about 100,000 data sets were built from the initial data state of 10,000 data. The AI-generated dataset showed marginally higher AUC but also a higher BCE score and a lower F1 score compared to the original. These outcomes highlight GPT-4’s potential in increas- ing data volume and diversity, but they also underline the need for more precise prompts to better emulate learner behaviors and performance, particularly in the context of improving BCE and F1 scores.
Submission Number: 151
Loading