EKT: Exercise-aware Knowledge Tracing for Student Performance Prediction (1906.05658v1)

Abstract: For offering proactive services to students in intelligent education, one of the fundamental tasks is predicting their performance (e.g., scores) on future exercises, where it is necessary to track each student's knowledge acquisition during her exercising activities. However, existing approaches can only exploit the exercising records of students, and the problem of extracting rich information existed in the exercise's materials (e.g., knowledge concepts, exercise content) to achieve both precise predictions of student performance and interpretable analysis of knowledge acquisition remains underexplored. In this paper, we present a holistic study of student performance prediction. To directly achieve the primary goal of prediction, we first propose a general Exercise-Enhanced Recurrent Neural Network (EERNN) framework by exploring both student's records and the exercise contents. In EERNN, we simply summarize each student's state into an integrated vector and trace it with a recurrent neural network, where we design a bidirectional LSTM to learn the encoding of each exercise's content. For making predictions, we propose two implementations under EERNN with different strategies, i.e., EERNNM with Markov property and EERNNA with Attention mechanism. Then, to explicitly track student's knowledge acquisition on multiple knowledge concepts, we extend EERNN to an explainable Exercise-aware Knowledge Tracing (EKT) by incorporating the knowledge concept effects, where the student's integrated state vector is extended to a knowledge state matrix. In EKT, we further develop a memory network for quantifying how much each exercise can affect the mastery of students on concepts during the exercising process. Finally, we conduct extensive experiments on large-scale real-world data. The results demonstrate the prediction effectiveness of two frameworks as well as the superior interpretability of EKT.

• The paper introduces two frameworks, EERNN and EKT, that integrate exercise records with embedded text and knowledge concepts to predict student performance.
• The EERNN framework employs bidirectional LSTM and attention mechanisms to capture exercise semantics and model student learning sequences.
• The EKT framework refines prediction by modeling multi-concept knowledge states with a memory network, offering enhanced interpretability and accuracy.

Exercise-aware Knowledge Tracing for Student Performance Prediction

The paper presents an investigation into student performance prediction within intelligent education systems by introducing two novel frameworks: Exercise-Enhanced Recurrent Neural Network (EERNN) and Exercise-aware Knowledge Tracing (EKT). The research aims to improve the prediction of student performance on future exercises by leveraging not only students' exercise records but also the rich materials embedded in those exercises, such as text content and knowledge concepts.

EERNN Framework

EERNN is designed to enhance the prediction of student performance by integrating student exercise sequences and the corresponding text content of these exercises. It involves two primary components: Exercise Embedding and Student Embedding. The Exercise Embedding utilizes a bidirectional LSTM to generate semantic representations of exercises based on their text content, which allows differentiation of individual exercise characteristics automatically. The Student Embedding models the sequence of exercise attempts by embedding both the exercise representations and the obtained scores into a recurrent network.

For prediction, EERNN introduces two strategies:

• EERNNM: This employs the Markov property to predict the student’s next performance based only on the current state, capturing most recent learning influence.
• EERNNA: This utilizes an attention mechanism to emphasize particular historical insights when similar exercises are encountered, capitalizing on exercises with higher relevance to improve future performance predictions.

EKT Framework

To address limitations in interpretability and explicit knowledge tracing, the EKT framework extends EERNN by incorporating exercise-related knowledge concept information. EKT refines student knowledge modeling from a singular vector to a matrix capturing knowledge states across multiple concepts. The integration of a Memory Network helps quantify the effect of exercises on student mastery over different concepts.

EKT also offers two prediction methodologies:

• EKTM: Similar to EERNNM, applies the Markov property but tracks and utilizes knowledge states across multiple concepts for prediction.
• EKTA: Extends the attention mechanism from EERNNA to a multi-concept knowledge tracing, enhancing prediction by recognizing important knowledge state changes over time.

Empirical Evaluation and Results

The paper conducts extensive experimentation on a large-scale, real-world dataset, demonstrating that both frameworks, especially EKT, significantly outperform existing and baseline approaches in terms of prediction accuracy across various problem scenarios, including cold-start challenges. The EKT approach is shown to offer superior interpretability, making it particularly adept at tracking knowledge acquisition in a manner that informs students of their strengths and weaknesses across different concepts.

Implications and Future Work

This research has significant implications for personalized education systems by providing tools that not only predict outcomes accurately but also enrich the interpretability of student progress. The ability to track and explain changes in concept mastery offers practical benefits in guiding personalized learning pathways and interventions. Future explorations could focus on expanding the frameworks to account for subjective exercises, incorporating insights from educational psychology, or applying the models to broader educational contexts beyond mathematics.

In conclusion, the presented frameworks mark a meaningful advance in educational data mining, effectively bridging current gaps in student performance prediction and knowledge state interpretability. These advancements could greatly enhance adaptive learning systems and contribute to improved educational strategies and policies.