Abstract:Tracing a student's knowledge growth given the past exercise answering is a vital objective in automatic tutoring systems to customize the learning experience. Yet, achieving this objective is a non-trivial task as it involves modeling the knowledge state across multiple knowledge components (KCs) while considering their temporal and relational dynamics during the learning process. Knowledge tracing methods have tackled this task by either modeling KCs' temporal dynamics using recurrent models or relational dynamics across KCs and questions using graph models. Albeit, there is a lack of methods that could learn joint embedding between relational and temporal dynamics of the task. Moreover, many methods that count for the impact of a student's forgetting behavior during the learning process use hand-crafted features, limiting their generalization on different scenarios. In this paper, we propose a novel method that jointly models the relational and temporal dynamics of the knowledge state using a deep temporal graph memory network. In addition, we propose a generic technique for representing a student's forgetting behavior using temporal decay constraints on the graph memory module. We demonstrate the effectiveness of our proposed method using multiple knowledge tracing benchmarks while comparing it to state-of-the-art methods.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to accurately track students' knowledge growth in an automated tutoring system in order to customize the learning experience. Specifically, the paper aims to jointly model the temporal and relational dynamics of knowledge components (KCs), while taking into account the influence of students' forgetting behavior, so as to more accurately predict students' performance on future problems. ### Detailed Problem Description 1. **Modeling multi - dimensional knowledge states**: - The paper needs to handle the complex relationships between multiple knowledge components (KCs) and capture the changes in these relationships over time. 2. **Modeling of forgetting behavior**: - Students' knowledge states will change over time, especially due to the forgetting effect. Existing methods usually rely on hand - designed features to represent forgetting behavior, which limits the generalization ability of the model. 3. **Effectively capturing the relationship between questions and KCs**: - In the process of knowledge tracing, understanding the dependency relationship between questions and KCs is crucial for accurately predicting students' knowledge states. ### Solution Overview To solve the above problems, the paper proposes a new method - **Temporal Graph Memory Network (TGMN)**. TGMN achieves joint modeling of knowledge states in the following ways: - **Joint embedding of temporal and relational dynamics**: TGMN combines the advantages of deep temporal models and graph models and can simultaneously model the temporal and relational dynamics of knowledge states within the same framework. - **General representation of forgetting behavior**: A general - purpose technique based on temporal decay constraints is introduced, which can represent students' forgetting behavior without the need for hand - designed features. ### Specific Components of the Model 1. **Temporal Graph Key - Value Memory**: - Use a key - value pair structure to store the knowledge state of each KC, where the key represents the identity of the KC and the value represents its mastery level. 2. **Key - Value GCN (Graph Convolutional Network)**: - Extract relevant information from the temporal graph memory through graph convolution operations to generate a read vector, which is used to represent the long - term knowledge context related to the current problem. 3. **Sequence Context Memory Cell**: - Use a GRU (Gated Recurrent Unit) cell to capture the short - term knowledge context in the current answering session and generate a short - term context vector. 4. **Answer Prediction Head**: - Concatenate the long - term and short - term context vectors and input them into a feed - forward neural network to predict the probability that a student will answer a question correctly. 5. **Memory Update Module**: - Update the value embeddings in the temporal graph memory module according to the prediction results and the relevance of the questions, and at the same time introduce a temporal decay mechanism for forgetting behavior. ### Experimental Verification The paper verifies the effectiveness of the proposed method through multiple knowledge - tracing benchmark datasets and compares it with existing state - of - the - art methods. The experimental results show that TGMN performs well on multiple datasets, especially having significant advantages in dealing with forgetting behavior and the temporal dynamics of knowledge states. In summary, through the introduction of the TGMN model, this paper effectively solves the problems of multi - dimensional modeling and forgetting behavior representation in knowledge tracing, providing strong support for personalized learning in automated tutoring systems.

Temporal Graph Memory Networks For Knowledge Tracing

Deep Graph Memory Networks for Forgetting-Robust Knowledge Tracing

DyGKT: Dynamic Graph Learning for Knowledge Tracing

A Temporal Convolutional Knowledge Tracing Model Integrating Forgetting Factors and Item Response Theory

Dynamic Key-Value Memory Networks With Rich Features for Knowledge Tracing

Knowledge Tracing with Sequential Key-Value Memory Networks

Recurrent knowledge tracing machine based on the knowledge state of students

Knowledge Relation Rank Enhanced Heterogeneous Learning Interaction Modeling for Neural Graph Forgetting Knowledge Tracing

A Deeper Knowledge Tracking Model Integrating Cognitive Theory and Learning Behavior

Temporal knowledge graph reasoning triggered by memories

Automatical Graph-based Knowledge Tracing

Time Interval Aware Self-Attention approach for Knowledge Tracing?

Learning consistent representations with temporal and causal enhancement for knowledge tracing

Leveraging Pedagogical Theories to Understand Student Learning Process with Graph-based Reasonable Knowledge Tracing

Deep Knowledge Tracing with Side Information

Sthkt: Spatiotemporal Knowledge Tracing with Topological Hawkes Process

DKVMN-KAPS: Dynamic Key-Value Memory Networks Knowledge Tracing With Students' Knowledge-Absorption Ability and Problem-Solving Ability

DKVMN&MRI: A New Deep Knowledge Tracing Model Based on DKVMN Incorporating Multi-Relational Information

Heterogeneous Graph Representation for Knowledge Tracing.

History Repeats: Overcoming Catastrophic Forgetting For Event-Centric Temporal Knowledge Graph Completion

DGEKT: A Dual Graph Ensemble Learning Method for Knowledge Tracing