https://arxiv.org/abs/2007.12324

Aritra Ghosh, Neil Heffernan, and Andrew S Lan. 2020. Context-aware attentive knowledge tracing. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2330–2339.

University of Michigan - Ann Arbor

 - Introduction
 - Knowledge Tracing Problem Setup
 - The AKT Method
 - Experiment
 - Conclusion

Context-Aware Attentive Knowledge Tracing

Context-Aware Attentive Knowledge Tracing (Reference)

The paper proposes the attentive knowledge tracing (AKT) method, which uses a series of attention networks to draw connections between this question and every question the learner has responded to in the past. 

The key innovations include:
 - Instead of use raw question and response embeddings, the paper put raw embeddings into context and use context-aware representations of past questions
and responses by taking a learner’s entire practice history into account.
 - The paper proposes a novel monotonic attention mechanism that uses an exponential decay curve to down weight the importance of questions in the distant past. It also develop a context-aware measure to characterize the time distance between questions a learner has responded to in the past.
 - Leveraging a series of Rasch model-based embeddings to capture individual differences among questions.

Introduction (Context-Aware Attentive Knowledge Tracing)

For learner $i$ at time step $t$ , we denote the combination of the question that they answered, the concept this question covers, and their graded response as a tuple, ($q_t^i, c_t^i, r_t^i$), where $q_t^i \in N^+$ is the question index, $c_t^i \in N^+$ is the concept index, and $r_t^i$ ∈ {0, 1} is the response.

Given a learner's past history up to time $t − 1$ as $\{(q_1, c_1, r_1), . . . , (q_{t−1}, c_{t−1}, r_{t−1})\}$, the paper aims to predict their response $r_t$ to question $q_t$ on concept $c_t$ at the current time step, $t$.

The paper uses real-valued embedding vectors $x_t ∈ R^D$ and $y_t ∈ R^D$ to represent each question and each question-response pair $(q_t , r_t )$, respectively. $x_t$ characterizes information about questions, and $y_t$ characterizes the knowledge learners acquire by responding to questions, with two separate embeddings for correct and incorrect responses, respectively.

Knowledge Tracing Problem Setup (Context-Aware Attention Knowledge Tracing)

The AKT method consists of four components: two self-attentive encoders, one for questions and one for knowledge acquisition, a single attention-based knowledge retriever, and a feed-forward response prediction model.

The Rasch model-based embeddings are used as raw embeddings for questions and responses. The question and knowledge encoders compute the context-aware representations of questions and responses pairs. The knowledge retriever uses these representations as input and computes the knowledge state of the learner.

Learn Before

Related