RKT : Relation-Aware Self-Attention for Knowledge Tracing

Shalini Pandey,Jaideep Srivastava

DOI: https://doi.org/10.1145/3340531.3411994

2020-08-29

Abstract:The world has transitioned into a new phase of online learning in response to the recent Covid19 pandemic. Now more than ever, it has become paramount to push the limits of online learning in every manner to keep flourishing the education system. One crucial component of online learning is Knowledge Tracing (KT). The aim of KT is to model student's knowledge level based on their answers to a sequence of exercises referred as interactions. Students acquire their skills while solving exercises and each such interaction has a distinct impact on student ability to solve a future exercise. This \textit{impact} is characterized by 1) the relation between exercises involved in the interactions and 2) student forget behavior. Traditional studies on knowledge tracing do not explicitly model both the components jointly to estimate the impact of these interactions. In this paper, we propose a novel Relation-aware self-attention model for Knowledge Tracing (RKT). We introduce a relation-aware self-attention layer that incorporates the contextual information. This contextual information integrates both the exercise relation information through their textual content as well as student performance data and the forget behavior information through modeling an exponentially decaying kernel function. Extensive experiments on three real-world datasets, among which two new collections are released to the public, show that our model outperforms state-of-the-art knowledge tracing methods. Furthermore, the interpretable attention weights help visualize the relation between interactions and temporal patterns in the human learning process.

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

The paper attempts to address the problem of how to more accurately model the changes in students' knowledge states over a series of exercises in Knowledge Tracing (KT). Specifically, the paper focuses on two key factors:

1. **Relationships between exercises**: The relationships between different exercises can affect students' performance on new exercises. For example, if two exercises belong to the same knowledge point, a student's performance on one exercise will directly influence their performance on the other.
2. **Students' forgetting behavior**: After completing an exercise, students may forget the learned knowledge over time, which will also impact their performance on subsequent exercises.

Traditional knowledge tracing methods usually do not consider these two factors simultaneously, thus having certain limitations in predicting students' future performance. To address this issue, the paper proposes a new Relation-aware Self-Attention model for Knowledge Tracing (RKT). This model integrates the relationship information between exercises and the information on students' forgetting behavior through the introduction of a relation-aware self-attention layer, thereby more accurately predicting students' performance on the next exercise.