Findings suggest that closed-loop coordination among state representation, policy optimization, and educational constraints contributes to improvements in personalized learning path recommendation quality.
Abstract
Smart education platforms accumulate large volumes of learning resources and interaction logs. However, fixed recommendation sequences are often unable to adapt to differences in knowledge foundations, learning behaviors, and cognitive progress among learners. Existing learning path recommendation methods primarily focus on static matching and single-step ranking, while providing limited support for modeling learning state evolution and long-term path benefits. To improve learning resource recommendation accuracy, path continuity, and learning completion outcomes, a personalized learning path (PLP) optimization model integrating Transformer and Reinforcement Learning is developed. The model aims to generate learning resource sequences that align with individual cognitive processes through dynamic state awareness, sequential policy decision-making, and the incorporation of educational constraints. First, learning behavior sequences, knowledge point information, answer feedback, resource access records, and time intervals are encoded within a unified framework. A Transformer is employed to capture long-range dependencies and dynamic state features. Second, PLP generation is formulated as a continuous decision-making process, in which RL performs policy search within the candidate resource space. Finally, knowledge prerequisite relationships, difficulty progression rules, learning load boundaries, and a multi-objective reward function are incorporated into a unified optimization framework. Offline simulation experiments are conducted on three public datasets, namely EdNet, ASSISTments, and Junyi. Performance is compared with Self-Attentive Knowledge Tracing, Separable Self-Attentive Neural Knowledge Tracing Plus, Unified Knowledge Tracing (UniKT), and Time-Aware Reinforcement Learning. Experimental results indicate that Transformer and Reinforcement Learning for Learning Path (TFRL-Path) achieves a hit ratio of 0.846 on EdNet, exceeding UniKT by 0.025. On ASSISTments, the mean average precision reaches 0.737 and the completion rate reaches 0.801. On Junyi, precision reaches 0.736 and convergence epochs decrease to 38. The cross-dataset average composite score reaches 0.798 with a standard deviation of 0.028. The prerequisite satisfaction rate reaches 0.872, while the resource repetition rate decreases to 0.121. These findings suggest that closed-loop coordination among state representation, policy optimization, and educational constraints contributes to improvements in personalized learning path recommendation quality.
Personalized learning aims to tailor learning content to individual learners based on their performance, conduct, and interests. Conventional approaches lack adaptability in dynamic decision-making and don’t handle uncertainty in learner behaviour, leading to less-than-optimal learning paths. This research aims to develop a Personalized Optimization Engine for Reinforced Sequencing of Japanese Learning Activities (PERSOJAPAN), capable of adapting to individual learner profiles and providing adaptive personalized learning performance. To overcome these shortfalls, introduce a proposed method, an intelligent framework integrating Partially Observable Markov Decision Process (POMDP), Duelling Deep Q-Network (DDQN), and Synergistic Ant Colony Optimization (SACO) for optimal learning path personalization. The model is learned and tested with the Japan Learning Dataset from Kaggle containing learner activity and performance information. The dataset is pre-processed with Z-score normalization and then feature extracted with Principal Component Analysis (PCA) to improve computational efficiency. POMDP models the probabilistic aspect of learner states, DDQN discovers optimal learning actions by applying deep reinforcement learning, and SACO optimizes the path personalization with bio-inspired optimization. This combined framework facilitates adaptive, dynamic learning that is appropriate for every student’s changing requirements. Performance is measured through recall of 86.1%, F1-score of 87.4%, and AUC of 0.978, wherein the designed method performs better than current baselines, reflecting enhanced accuracy, decision-making quality, and personalization strength. Overall, the framework successfully addresses the limitations of traditional rule-based and static systems by introducing a scalable, data-driven approach that adapts to individual learner needs, enhances engagement, supports adaptive personalized learning feedback, and continuously evolves to improve personalized learning outcomes effectively.
Intelligent recommendations are becoming more and more popular in Chinese language learning (CLL) platforms, yet current approaches lack centralization, slowness in adapting to the learners, and poor interaction modeling. The classical forms of Reinforcement Learning (RL) are typically not effective in providing individualized recommendations of resources in the context of dynamic learning. As a way of resolving these limitations, the focus is on developing a flexible and decentralized recommendation system that is efficient in responding to diverse learner profiles and changing patterns of engagement. A new Decentralized Distributed Proximal using Dueling Deep Q Network (D2P-D2QN) is presented, which combines the accuracy of the D2QN estimation with the robustness of proximal policy optimization in a multi-agent setting that is distributed. A Chinese Learning Resource dataset containing 2000 records of learner performance logs, resource metadata, engagement durations, and feedback indicators supports this system. Preprocessing includes missing-value correction for data integrity, normalization for stable learning, and sequence structuring to model temporal learner behavior. Feature extraction relies on an autoencoder to generate compact latent representations of learner-resource interactions. Within the proposed mechanism, the proximal component manages policy refinement, while the dueling Q network handles advantage estimation and value decomposition. Decentralized distributed coordination enables multiple agents to update decisions concurrently for scalable and responsive recommendation generation. Implementation utilizes a deep learning environment supported by Python and RL libraries. Experimental results demonstrate 95.6% accuracy, 93.2% F1-score, 0.98 AUC, 0.94 MCC, 160 ms runtime, 72 ms latency, learning willingness 4.5, engagement 4.6, and 94.1% interaction success compared to conventional RL-based recommenders. The overall framework indicates strong potential for next-generation adaptive educational systems that require both precision and scalability in resource delivery.
Ming Li· Discover Artificial Intellig...· 0 citations
Deep reinforcement learning offers a robust solution to the challenge of adaptive learning path optimization in higher education, where student heterogeneity and curriculum complexity often undermine traditional instructional models. This paper introduces a technically advanced recommendation framework for university-level Chinese language courses, which models student knowledge states and behavioral data as a Markov Decision Process and applies a deep Q-network to predict optimal content sequencing. The system monitors progress, time efficiency and participation thru a multidimensional reward mechanism to achieve teaching goals. The experiment collected a large amount of interaction data between students and other students. The proposed method is compared with the existing rule-based system. The findings show that the accuracy and efficiency of the customized curriculum are greatly improved. The accuracy of learning tasks increased by 20%, and the average completion time decreased by 30%. The modular architecture allows for scalable deployment, efficient data processing, and precise feature engineering in heterogeneous educational environments. Limitations are principally related to the dependency on high-quality annotated data and the computational demands of model training, although the system exhibits significant robustness and generalizability within the tested context. These results show that deep reinforcement learning is useful in intelligent education systems. It provides a reference point for future technological development and encourages adaptive teaching.
Liqun Fang· International Conference on...· 0 citations
Large language models used in personalized education often face mismatches between generated learning paths and teaching objectives, limited adaptive generation ability, and insufficient modeling of heterogeneous learner cognitive profiles. This paper proposes an adaptive alignment method for learning path generation based on the IB-GRPO large language model. First, a multidimensional feature space is constructed by integrating a knowledge graph, learner profiles, and teaching objectives. A knowledge-node importance assessment module dynamically assigns weights to learning content. Second, learners are clustered according to cognitive levels, and a group relative policy optimization algorithm updates path strategies within each group. Third, a teaching-objective constraint function is introduced, and reinforcement learning is used to realize adaptive alignment between generated paths and learning objectives. Finally, a feedback loop dynamically adjusts path-generation strategies according to learning results. Experiments on a dataset of 5,000 learners show that the method achieves a target coverage rate of 90.8% and an ability matching rate of 93.1%. The results demonstrate that the grouping strategy effectively distinguishes learners’ needs and improves personalized path generation.
Generating personalized learning paths remains an open problem in intelligent education systems, where conventional sequence recommendation methods typically output ranked resource ID lists without encoding pedagogical intent or supporting post-deployment refinement. To address these limitations, we present DLPP (Diffusion-based Learning Path Planning Framework), which recasts path generation as a two-stage process of instructional planning followed by semantic instantiation. In the first stage, a conditional diffusion model produces structured activity-type sequences within a continuous embedding space; nearest-neighbor quantization then maps these sequences to discrete pedagogical categories. In the second stage, a Large Language Model (LLM) equipped with Retrieval-Augmented Generation (RAG) converts each abstract plan step into a concrete, resource-grounded learning activity. An ensemble of five student behavior simulators supplies uncertainty-aware quality scores, and a Diffuser-based reinforcement learning module closes the optimization loop. Evaluated on two public educational datasets, DLPP yields 7–8% PKG improvement over the pure diffusion baseline on both EdNet and Junyi Academy. On the medium-scale EdNet (2941 training paths), it outperforms RL-Path and DiffPath baselines while remaining competitive with rule-based methods.
Yuan Ren, Zhanfang Chen, Zeming Du et al.· International Conference on...· 0 citations
A reinforcement learning framework is developed for adaptive learning path recommendation in large-scale English MOOCs, leveraging a Deep Q-Network (DQN) architecture integrated with unsupervised clustering. K-means++ initializes and optimizes strategies by extracting consistent user groups, and unsupervised clustering analysis addresses data sparsity and cold start scenarios. The recommendation strategy is updated in real-time thru a composite reward function, which is sensitive to completion rates, engagement metrics, and evaluation gains. Comprehensive experiments using the OULAD and proprietary MOOC datasets indicate that, compared to traditional recommendation baselines, the proposed method has achieved significant improvements in accuracy, recall, and F1-score. Robustness tests confirmed the model's resilience to noise and incomplete data. This indicates that the model maintains adaptability in diverse learning pathways. These findings demonstrate the effectiveness and scalability of combining DQN-based reinforcement learning with advanced student profiling in a MOOC environment.
Xin Zhang, Mei Li, Yujiao Han· International Conference on...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
