BamaER: A Behavior-Aware Memory-Augmented Model for Exercise Recommendation

TL;DR

BamaER addresses the challenge of personalized exercise recommendation by moving beyond plain exercise sequences to incorporate rich behavioral interactions and long-term knowledge trajectories. It introduces three components: a learning progress predictor with tri-directional hybrid encoding, a memory-augmented knowledge tracing module for stable mastery estimation, and a diversity-focused exercise filter aided by the Hippopotamus Optimization Algorithm. Across five real-world datasets, BamaER achieves superior accuracy and diversity compared with eight baselines, with ablation studies confirming the value of each module. The framework demonstrates how behavior-informed progress signals and memory-aware mastery can improve both the relevance and coverage of recommended exercises, offering practical benefits for intelligent education systems.

Abstract

Exercise recommendation focuses on personalized exercise selection conditioned on students' learning history, personal interests, and other individualized characteristics. Despite notable progress, most existing methods represent student learning solely as exercise sequences, overlooking rich behavioral interaction information. This limited representation often leads to biased and unreliable estimates of learning progress. Moreover, fixed-length sequence segmentation limits the incorporation of early learning experiences, thereby hindering the modeling of long-term dependencies and the accurate estimation of knowledge mastery. To address these limitations, we propose BamaER, a Behavior-aware memory-augmented Exercise Recommendation framework that comprises three core modules: (i) the learning progress prediction module that captures heterogeneous student interaction behaviors via a tri-directional hybrid encoding scheme; (ii) the memory-augmented knowledge tracing module that maintains a dynamic memory matrix to jointly model historical and current knowledge states for robust mastery estimation; and (iii) the exercise filtering module that formulates candidate selection as a diversity-aware optimization problem, solved via the Hippopotamus Optimization Algorithm to reduce redundancy and improve recommendation coverage. Experiments on five real-world educational datasets show that BamaER consistently outperforms state-of-the-art baselines across a range of evaluation metrics.

Figures (5)

• Figure 1: Illustration of key challenges. (a) Learning progress depends on response accuracy: perfect scorers advance faster than those who make mistakes. Excluding this, models may equate all students’ progress. (b) Existing methods’ non-interactive fixed-length subsequence splits lose long-range dependencies, limiting short-term context models’ accuracy in estimating students’ current knowledge.
• Figure 2: Overall architecture of BamaER. (a) We take the student’s historical exercise sequence as input, which is arranged in the order of interactions and includes the knowledge concept of each exercise together with the corresponding response outcome. (b) We then input the sequence into the learning progress prediction module to and the knowledge concept mastery prediction module. These two modules embed features such as exercises, knowledge concepts, and scores into the model, thereby generating the learning progress and the knowledge concept mastery for each student. (c) These two matrices are input into the exercise filter module, where they serve as criteria for selecting a candidate set from the exercise repository. The candidate set is then optimized using the ER-HO algorithm, ultimately generating the exercise recommendation list.
• Figure 3: Comparison of accuracy and diversity metrics under different numbers of exercise recommendations on five datasets.
• Figure 4: Comparison of accuracy and diversity metrics under different numbers of population on five datasets.
• Figure 5: Detailed comparison of the performance of the exercise recommendation list across three key metrics on ASSISTments datasets.