Improving Knowledge Tracing via Pre-training Question Embeddings

Abstract

<jats:p>Knowledge tracing (KT) defines the task of predicting whether students can correctly answer questions based on their historical response. Although much research has been devoted to exploiting the question information, plentiful advanced information among questions and skills hasn't been well extracted, making it challenging for previous work to perform adequately. In this paper, we demonstrate that large gains on KT can be realized by pre-training embeddings for each question on abundant side information, followed by training deep KT models on the obtained embeddings. To be specific, the side information includes question difficulty and three kinds of relations contained in a bipartite graph between questions and skills. To pre-train the question embeddings, we propose to use product-based neural networks to recover the side information. As a result, adopting the pre-trained embeddings in existing deep KT models significantly outperforms state-of-the-art baselines on three common KT datasets.</jats:p>

Knowledge structure enhanced graph representation learning model for attentive knowledge tracing

“…4 For this reason, KT has been a popular interdisciplinary research topic across education, computer science, and cognitive science. 4,6,13 Massive efforts have been devoted to skill-level KT, 12,14 which performs KT based on the skills (or "knowledge concepts") required in a specific domain. Each question in a KT task is correlated with one or more skills needed to solve the question (e.g., "3+5" corresponds to the skill "addition of integers"), and each skill is related to many questions.…”

“…Accordingly, the response data are quite sparse. 12,14 Skill-level KT is feasible to some extent because skill mastery can largely affect the correctness of question answering. Therefore various KT methods proposed under this setting have achieved good performance.…”

“…This approach loses the distinctive information related to individual questions, leading to imprecise inferences of the learners' knowledge states. 12,[14][15][16] For example, in Figure 1(left), the questions "3+5" and "345+6789" both require the skill "addition of integers," and are considered as the same inputs when building the KT models, which ignores their different difficulty levels. Existing research have proved that question difficulty undoubtedly influences the learner performance, 11,17,18 and the relative difficulty level of a specific question varies from learner to learner.…”

A Dynamic Knowledge Diagnosis Approach Integrating Cognitive Features

“…In practice, the main task in such educational platforms is to predict student performance [7]; that is, to forecast whether a student can answer exercises (e.g., e7, e8) correctly in a future assessment. Meanwhile, it also requires tracking the change in students' knowledge mastery level [8] (K1, K2, and K3) in their assessment process.…”

“…On the other hand, the exercise consisting of the same knowledge may have different latent features [8], [19], [20], such as difficulty. In fact, most existing knowledge tracing methods, such as Bayesian knowledge tracing (BKT) [21] and DKT [22], usually use the corresponding knowledge instead of exercises themselves, ignoring important latent exercise features in their exercise process.…”