Pith. sign in

REVIEW 5 major objections 6 minor 46 references

MAS-KCL: Knowledge component graph structure learning with large language model-based agentic workflow

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read MAS-KCL uses LLM agents to adaptively steer differential evolution for knowledge component graph learning and reports lower loss than four baselines on nine datasets.

desk verdict A plausible LLM-agent plus DE extension for KC graph learning held back by an undefined loss metric and weak statistics. read the letter →

arxiv 2505.14126 v1 pith:MHMCJUIU submitted 2025-05-20 cs.LG cs.CYcs.HCcs.MA

Yuan-Hao Jiang , Kezong Tang , Zi-Wei Chen , Yuang Wei , Tian-Yi Liu , Jiayi Wu This is my paper · ORCID
classification cs.LGcs.CYcs.HCcs.MA
keywords graphlearningalgorithmknowledgelearnersstructureachievedatasets
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Knowledge components are the small building blocks of a subject, like 'addition' or 'multiplication'. A knowledge component graph is a map showing which blocks must be learned before others. This paper describes a program, MAS-KCL, that tries to build such maps automatically from student data.

The program starts with many random candidate graphs. It uses an evolutionary algorithm, differential evolution, to combine and mutate them. What is new is a team of large language model agents that watch how the search is going. One agent, the 'game agent', decides whether the search should focus on refining good graphs or exploring new ones. Two other agents look at edges in the graphs: one encourages edges that seem to help, and the other discourages edges that seem to hurt. These agents hand their decisions to the algorithm, which then adjusts the search.

The authors tested MAS-KCL on nine datasets, five synthetic and four real. On all nine, it reported a lower loss value than four older optimization algorithms. The ablation removes each agent one at a time and shows that loss gets worse, suggesting the agents are doing useful work.

The main caveat is that the loss metric itself comes from the authors' earlier paper, and the results have no statistical significance tests. The paper does not release code or the exact prompts used to talk to the language models, so another lab cannot easily check whether the improvement is real or reproducible. If the method does work, teachers could get more reliable prerequisite maps to plan instruction, but the evidence in this paper is not yet strong enough to back that conclusion.

Extended reading notes

Core claim

The proposed MAS-KCL achieves the best results in every case, demonstrating its effectiveness in the task of KC graph structure learning (Section 4.2). If correct, MAS-KCL produces more accurate knowledge component graphs than the four compared baselines as measured by the loss metric inherited from reference [20].

Load-bearing premise

The loss metric (1 - F) inherited from the authors' prior paper [20] is a valid measure of true learning path recognition quality. This is assumed throughout (Section 3.2: 'fitness definition aligns with prior work [20]'; Section 4.1.3: 'Following previous work, we employed the loss metric to evaluate the performance [20]'). If this metric does not track real prerequisite relationships, all reported loss improvements are irrelevant to the stated educational goal.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. The paper proposes MAS-KCL, a multi-agent LLM-driven evolutionary algorithm for knowledge component (KC) graph structure learning. The method combines differential evolution with a dynamic population split and a bidirectional feedback mechanism controlled by Game, Positive Feedback, and Negative Feedback Agents, all implemented via LLM prompts. Experiments on five synthetic and four real-world educational datasets compare MAS-KCL against four baselines using a loss metric inherited from prior work [20], and the paper reports that MAS-KCL achieves the best loss in every case. Ablation, LLM-sensitivity, and convergence experiments are also presented. The central claim is that MAS-KCL improves the accuracy of learning-path recognition, but the evaluation relies on an undefined metric and on comparisons without statistical support.

Significance. If the loss metric were properly defined and validated, MAS-KCL would be a useful contribution at the intersection of LLM-based agent systems and educational graph structure learning. The paper has several strengths: it evaluates on multiple real-world datasets, includes ablation studies for each agent, reports a convergence analysis with 30 runs, and compares against four baselines. However, the central claim of universal best performance is not self-contained because the loss metric is never defined, and the statistical evidence is insufficient. The current manuscript does not establish that the reported loss improvements correspond to more accurate KC graphs or better learning-path recognition, so the significance of the empirical results remains uncertain.

major comments (5)
  1. [Section 3.2 and Section 4.1.3] The loss metric is never defined. The paper only states that "the fitness definition aligns with prior work [20]" and "Following previous work, we employed the loss metric to evaluate the performance [20]", but it does not specify the functional form of F, how the ground-truth graph is represented for each real-world dataset, or how the loss is computed. This is load-bearing because the central claim in Section 4.2 that "MAS-KCL achieves the best results in every case" is measured exclusively by this metric, and for MOOCCubeX-Math the ground truth consists of "partially annotated and algorithmically predicted prerequisite relationships" (Section 4.1.1). The manuscript needs to define F and the exact ground-truth edges used, and provide an external validation (e.g., agreement with expert-annotated prerequisites or improvement in a downstream learning-outcome prediction task) showing that lower loss corresponds to more accurate learning paths.
  2. [Section 4.2, Table 2] The claim of universal best performance is not statistically supported. The text says each algorithm was run "at least three times", but no exact run counts, significance tests, or effect sizes are reported. For MOOCCubeX-Math, MAS-KCL (2.51e-1 ± 7.61e-2) and EESHHO (2.57e-1 ± 4.95e-2) differ by 0.006, well within one standard deviation of either method, so the reported difference is not meaningful without additional evidence. Paired significance tests (e.g., Wilcoxon signed-rank) and per-run results are needed before claiming superiority on every dataset.
  3. [Section 4.3, Table 3 and Section 4.4, Table 4] The generalization and LLM-sensitivity claims are based on point estimates without adequate uncertainty quantification. Table 3 reports means and standard deviations for MAS-KCL and an unnamed "Baseline", but no per-dataset repeats or significance tests, and Table 4 reports no standard deviations at all. Consequently, the claims that "MAS-KCL reduces the loss by 5.51%" and "GPT models reduce the loss by 5.79%" are not supported. In addition, the "∆loss" values in Table 3 appear to be arithmetic differences in percentage units (e.g., 32.81 - 26.99 = 5.82) rather than relative percentage reductions; this should be clarified, and the baseline should be explicitly identified as EESHHO.
  4. [Section 4.5, Table 5 and Figure 4] The ablation results are presented as single mean values with no repeated-run statistics. For example, the conclusion that "removal of any single agent led to an increase in loss" and that the Negative Feedback Agent plays the most critical role is based on differences of roughly 0.01 in loss across datasets, but without significance tests or confidence intervals it is not possible to determine whether the observed ordering of ablation effects is reliable. The manuscript should include repeated runs and appropriate significance testing for all ablation comparisons.
  5. [Algorithm 1 and Section 3.2] Algorithm 1 is not sufficiently specified for replication. Lines 7-15 describe a "bidirectional feedback mechanism" using "count ones", "Positive Factor", and "Negative Factor", but the precise edge-vector operations, the initialization and update rules for PF and NF, and the mapping from LLM decisions to parameter changes are not defined. The prompt templates and LLM sampling settings for the Game Agent and Feedback Agents are omitted, and Table 1 lists only PF=0.6, NF=0.4, AP=0.4 without ranges or update rules. A supplementary appendix with the exact update equations, prompt templates, and LLM configuration (including temperature and seeds) is required.
minor comments (6)
  1. [Figure 1] The figure contains the typo "Agentic Wrokflow"; it should read "Agentic Workflow".
  2. [Table 3] The caption says "Baseline" but does not identify which algorithm is used; based on Section 4.3 it should be EESHHO, the best-performing baseline from Section 4.2.
  3. [Figure 5] The y-axis label "loss (%)" is inconsistent with the fractional loss values reported in Table 2; please unify the units across tables and figures.
  4. [Table 4] The model name "GPT-4.0" is non-standard; please specify the exact model versions and access dates used for GPT-4.0 and GPT-3.5.
  5. [Section 4.1.1] The name "LPR-RWD" is used both for the series of real-world datasets and for a specific dataset in that series; please disambiguate these names to avoid confusion.
  6. [Data Availability] The data availability statement says datasets are accessible "upon request" but provides no repository URL or direct links; please include the actual access links for each dataset.
Assumptions & free parameters 6 free parameters · 4 assumptions · 3 invented entities

These parameters and assumptions carry the load of the central claim. The hand-set factors AP, PF, and NF directly control the search; population size and maxFE are unstated; the loss metric comes from the authors' prior work; dataset labels and LLM decision stability are assumed; and the three AI agents have internal ablation support only.

free parameters (6)
  • AP_initial (Ambient Pressure) = 0.4
    Initial share of elite individuals retained each iteration; set by hand in Table 1, then adjusted by the Game Agent during search.
  • PF (Positive Factor) = 0.6
    Weight applied to newly added edges in the positive feedback mechanism; set by hand in Table 1.
  • NF (Negative Factor) = 0.4
    Weight applied to newly added edges in the negative feedback mechanism; set by hand in Table 1.
  • Population size N
    Required by Algorithm 1 but not reported in the paper; the reported results depend on it.
  • Maximum function evaluations maxFE
    Termination condition in Algorithm 1; not reported.
  • LLM prompt templates and sampling settings
    The agent decisions come from LLM calls with unspecified prompts, temperatures, and seeds; these settings affect the loss values.
assumptions (4)
  • domain assumption The loss function L = 1 - F from reference [20] is a valid measure of KC graph quality and learning path recognition accuracy.
    Used as the sole evaluation metric (Section 4.1.3) and as the fitness to optimize (Section 3.2). If this metric does not track real prerequisite relationships, the experimental claims do not follow.
  • domain assumption The ground-truth prerequisite relationships in LPR-RWD and MOOCCubeX-Math are reliable annotations of true KC dependencies.
    The reported losses assume these datasets provide meaningful targets; the paper does not assess label noise or annotation errors.
  • domain assumption LLM agent decisions are stable and beneficial enough to serve as algorithmic controls.
    The agents are invoked with unspecified prompts and no seeds; the method assumes their outputs are consistent and improve the search across runs.
  • domain assumption Differential evolution operators on the decision-vector encoding produce valid KC graph structures.
    Algorithm 1 applies DE and edge addition/deletion to a decision vector without describing how acyclicity or graph validity is maintained.
invented entities (3)
  • Game Agent
    purpose: LLM-based controller that decides whether and how much to adjust the Ambient Pressure (AP) parameter, balancing exploration and exploitation.
    Only in-paper ablation evidence (removing it raises loss by 2.30%, Table 5). No external falsifiable handle is provided.
  • Positive Feedback Agent
    purpose: Identifies edges associated with decreased loss and promotes their addition across the population.
    Only in-paper ablation evidence (removing it raises loss by 2.80%, Table 5). No external replication.
  • Negative Feedback Agent
    purpose: Identifies edges associated with increased loss and suppresses their appearance in the population.
    Only in-paper ablation evidence (removing it raises loss by 3.03%, Table 5). No external falsifiable handle.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MAS-KCL: Knowledge component graph structure learning with large language model-based agentic workflow." pith.science (2026). https://pith.science/paper/MHMCJUIU

@misc{pith2026250514126,
  author       = {Pith},
  title        = {Pith review of: MAS-KCL: Knowledge component graph structure learning with large language model-based agentic workflow},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MHMCJUIU}},
  note         = {Machine review of arXiv:2505.14126}
}
read the original abstract

Knowledge components (KCs) are the fundamental units of knowledge in the field of education. A KC graph illustrates the relationships and dependencies between KCs. An accurate KC graph can assist educators in identifying the root causes of learners' poor performance on specific KCs, thereby enabling targeted instructional interventions. To achieve this, we have developed a KC graph structure learning algorithm, named MAS-KCL, which employs a multi-agent system driven by large language models for adaptive modification and optimization of the KC graph. Additionally, a bidirectional feedback mechanism is integrated into the algorithm, where AI agents leverage this mechanism to assess the value of edges within the KC graph and adjust the distribution of generation probabilities for different edges, thereby accelerating the efficiency of structure learning. We applied the proposed algorithm to 5 synthetic datasets and 4 real-world educational datasets, and experimental results validate its effectiveness in learning path recognition. By accurately identifying learners' learning paths, teachers are able to design more comprehensive learning plans, enabling learners to achieve their educational goals more effectively, thus promoting the sustainable development of education.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 37 canonical work pages

  1. [20]

    Sustainability (2024)

    Liu, T.Y., Jiang, Y.H., Wei, Y., Wang, X., Huang, S., Dai, L.: Educational practices and algorithmic framework for promoting sustainable development in education by iden- tifying real-world learning paths. Sustainability (2024). doi:10.3390/su16166871

    work page doi:10.3390/su16166871 2024

  2. [1]

    In: Proceedings of the 16th Interna- tional Conference on Educational Data Mining, pp

    All` egre, O., Yessad, A., Luengo, V.: Discovering prerequisite relationships between knowledge components from an inter- pretable learner model. In: Proceedings of the 16th Interna- tional Conference on Educational Data Mining, pp. 490–496 (2023)

    work page 2023

  3. [2]

    Qualita- tive Report 17(1), 200–221 (2012)

    Berrett, B., Murphy, J., Sullivan, J.: Administrator insights and reflections: Technology integration in schools. Qualita- tive Report 17(1), 200–221 (2012). URL https://eric.ed. gov/?id=ej973041

    work page 2012

  4. [3]

    In: User Modeling 2005: 10th International Conference, UM 2005, Edinburgh, Scot- land, UK, July 24-29, 2005

    Carmona, C., Mill´ an, E., P´ erez-de-la Cruz, J.L., Trella, M., Conejo, R.: Introducing prerequisite relations in a multi- layered bayesian student model. In: User Modeling 2005: 10th International Conference, UM 2005, Edinburgh, Scot- land, UK, July 24-29, 2005. Proceedings 10, pp. 347–356. Springer (2005)

    work page 2005

  5. [4]

    The Visual Computer pp

    Chen, X., Yao, Y., Guan, Z., Li, C., Guan, J., Pu, J., Liu, R., Sheng, B., Yin, S., Qin, Y.: Dsts-gf: a dual-stream temporal– spatial transformer with gated fusion for the classification of obstructive sleep apnea. The Visual Computer pp. 1–13 (2025)

    work page 2025

  6. [5]

    User mod- eling and user-adapted interaction 12, 371–417 (2002)

    Conati, C., Gertner, A., Vanlehn, K.: Using bayesian net- works to manage uncertainty in student modeling. User mod- eling and user-adapted interaction 12, 371–417 (2002)

    work page 2002

  7. [6]

    User Modeling and User-Adapted Interaction 16, 403–434 (2006)

    Desmarais, M.C., Meshkinfam, P., Gagnon, M.: Learned stu- dent models with item to item knowledge structures. User Modeling and User-Adapted Interaction 16, 403–434 (2006)

    work page 2006

  8. [7]

    Asia Pacific Education Review 25(3), 595–609 (2024)

    Feng, Y.: Introduction to causal graphs for education re- searchers. Asia Pacific Education Review 25(3), 595–609 (2024). doi:10.1007/s12564-024-09980-3

    work page doi:10.1007/s12564-024-09980-3 2024

Show all 46 references
  1. [8]

    URL https://eedi.com/projects/neurips-2022

    Gong, W., Pawlowski, N., Smith, D., Wang, Z., Jennings, J., Barton, C., Woodhead, S., Zhang, C.: Instructions and guide: Causal insights for learning paths in education (2022). URL https://eedi.com/projects/neurips-2022

    2022

  2. [9]

    URL https://codalab.lisn.upsaclay.fr/ competitions/5626

    Gong, W., Pawlowski, N., Smith, D., Wang, Z., Jennings, J., Barton, C., Woodhead, S., Zhang, C.: Neurips 2022 causalml challenge: Causal insights for learning paths in ed- ucation (2022). URL https://codalab.lisn.upsaclay.fr/ competitions/5626

    2022

  3. [10]

    Education and Information Technologies 28(9), 11,401–11,422 (2023)

    He, X., Fang, J., Cheng, H.N.H., Men, Q., Li, Y.: Inves- tigating online learners’ knowledge structure patterns by concept maps: A clustering analysis approach. Education and Information Technologies 28(9), 11,401–11,422 (2023). doi:10.1007/s10639-023-11633-8

    2023 doi

  4. [11]

    Future Generation Computer Systems 97, 849–872 (2019)

    Heidari, A.A., Mirjalili, S., Faris, H., Aljarah, I., Mafarja, M., Chen, H.: Harris hawks optimization: Algorithm and appli- cations. Future Generation Computer Systems 97, 849–872 (2019). doi:10.1016/j.future.2019.02.028

    2019 doi

  5. [12]

    The Visual Com- puter 39(8), 3647–3659 (2023)

    Huang, S., Liu, X., Tan, T., Hu, M., Wei, X., Chen, T., Sheng, B.: Transmrsr: transformer-based self-distilled gener- ative prior for brain mri super-resolution. The Visual Com- puter 39(8), 3647–3659 (2023)

    2023

  6. [13]

    In: The 4th Work- shop on Mathematical Reasoning and AI at NeurIPS’24, pp

    Jiang, Y.H., Li, R., Wei, Y., Jia, R., Shao, X., Hu, H., Jiang, B.: Synchronizing Verbal Responses and Board Writing for Multimodal Math Instruction with LLMs. In: The 4th Work- shop on Mathematical Reasoning and AI at NeurIPS’24, pp. 46–59. Neural Information Processing Syste...

    2024

  7. [14]

    Global Chinese Confer- ence on Computers in Education, Chongqing, China (2024)

    Jiang, Y.H., Li, R., Zhou, Y., Qi, C., Hu, H., Wei, Y., Jiang, B., Wu, Y.: AI Agent for Education: von Neumann Multi-Agent System Framework. Global Chinese Confer- ence on Computers in Education, Chongqing, China (2024). doi:10.48550/arXiv.2501.00083

    2024 doi arXiv

  8. [15]

    IEEE Transactions on Learning Technologies 10(4), 450–462 (2017)

    K¨ aser, T., Klingler, S., Schwing, A.G., Gross, M.: Dynamic bayesian networks for student modeling. IEEE Transactions on Learning Technologies 10(4), 450–462 (2017)

    2017

  9. [16]

    Journal of Intelligent & Fuzzy Systems 40(2), 3289–3299 (2021)

    Li, R.: An artificial intelligence agent technology based web distance education system. Journal of Intelligent & Fuzzy Systems 40(2), 3289–3299 (2021). doi:10.3233/JIFS-189369

    2021 doi

  10. [17]

    In: Conference Proceedings of the 28th Global Chinese Conference on Computers in Educa- tion (GCCCE 2024), pp

    Li, R., Jiang, Y.H., Wang, Y., Hu, H., Jiang, B.: A Large Language Model-Enabled Solution for the Au- tomatic Generation of Situated Multiple-Choice Math Questions. In: Conference Proceedings of the 28th Global Chinese Conference on Computers in Educa- tion (GCCCE 2024), pp. 1...

    2024

  11. [18]

    Neurocomput- ing 419, 273–286 (2021)

    Li, S., Wang, H., Pan, R., Mao, M.: Memorypath: A deep reinforcement learning framework for incorporating memory component into knowledge graph reasoning. Neurocomput- ing 419, 273–286 (2021)

    2021

  12. [19]

    Information Processing & Management 62(1), 103,913 (2025)

    Li, X., Guo, S., Wu, J., Zheng, C.: An interpretable polyto- mous cognitive diagnosis framework for predicting examinee performance. Information Processing & Management 62(1), 103,913 (2025). doi:10.1016/j.ipm.2024.103913

    2025

  13. [21]

    ACM Comput

    Meng, L., Shao, Y., Yuan, L., Lai, L., Cheng, P., Li, X., Yu, W., Zhang, W., Lin, X., Zhou, J.: A Survey of Distributed Graph Algorithms on Massive Graphs. ACM Comput. Surv. 57(2), 27:1–27:39 (2024). doi:10.1145/3694966

    2024 doi

  14. [22]

    In: 2022 IEEE 38th International Conference on Data Engineering (ICDE), pp

    Meng, L., Yuan, L., Chen, Z., Lin, X., Yang, S.: Index-based Structural Clustering on Directed Graphs. In: 2022 IEEE 38th International Conference on Data Engineering (ICDE), pp. 2831–2844 (2022). doi:10.1109/ICDE53745.2022.00257

    2022

  15. [23]

    doi:10.48550/arXiv.2412.11488

    Meng, L., Yuan, L., Lin, X., Li, C., Wang, K., Zhang, W.: Counting Butterflies over Streaming Bipartite Graphs with Duplicate Edges (2024). doi:10.48550/arXiv.2412.11488

    2024 doi arXiv

  16. [24]

    Com- puters & Industrial Engineering 152, 107,050 (2021)

    Mohammadi-Balani, A., Dehghan Nayeri, M., Azar, A., Taghizadeh-Yazdi, M.: Golden eagle optimizer: A nature-inspired metaheuristic algorithm. Com- puters & Industrial Engineering 152, 107,050 (2021). doi:10.1016/j.cie.2020.107050

    2021

  17. [25]

    In: Proceedings of the Genetic and Evolutionary Computation Conference, pp

    Panichella, A.: An improved pareto front modeling algorithm for large-scale many-objective optimization. In: Proceedings of the Genetic and Evolutionary Computation Conference, pp. 565–573. ACM (2022). doi:10.1145/3512290.3528732

    2022

  18. [26]

    Ad- vances in neural information processing systems 28 (2015)

    Piech, C., Bassen, J., Huang, J., Ganguli, S., Sahami, M., Guibas, L.J., Sohl-Dickstein, J.: Deep knowledge tracing. Ad- vances in neural information processing systems 28 (2015)

    2015

  19. [27]

    The Visual Computer 41(1), 491–516 (2025)

    Senior, H., Slabaugh, G., Yuan, S., Rossi, L.: Graph neural networks in vision-language image understanding: a survey. The Visual Computer 41(1), 491–516 (2025)

    2025

  20. [28]

    Nature Biomedical Engineering pp

    Sheng, B., Keane, P.A., Tham, Y.C., Wong, T.Y.: Synthetic data boosts medical foundation models. Nature Biomedical Engineering pp. 1–2 (2025)

    2025

  21. [29]

    Expert Systems with Applications 263(2025), 125,723 (2025)

    Shi, J., Zhao, J., Wu, X., Xu, R., Jiang, Y.H., He, L.: Mitigating Reasoning Hallucination Through Multi-Agent Collaborative Filtering. Expert Systems with Applications 263(2025), 125,723 (2025). doi:10.1016/j.eswa.2024.125723 MAS-KCL: Knowledge Component Graph Structure Learn...

    2025

  22. [30]

    Machine Learning and Knowledge Extraction 6(4), 2722–2737 (2024)

    Shokouhinejad, H., Razavi-Far, R., Higgins, G., Ghorbani, A.A.: Node-centric pruning: A novel graph reduction ap- proach. Machine Learning and Knowledge Extraction 6(4), 2722–2737 (2024)

    2024

  23. [31]

    Innovations in Ed- ucation and Teaching International 43(2), 109–120 (2024)

    Spector, J.M.: A methodology for assessing learning in com- plex and ill-structured task domains. Innovations in Ed- ucation and Teaching International 43(2), 109–120 (2024). doi:10.1080/14703290600650368

    2024 doi

  24. [32]

    The Visual Computer pp

    Sun, L., Yan, J., Qiu, Y., Qin, W.: The crowd cooperation approach for formation maintenance and collision avoidance using multi-agent deep reinforcement learning. The Visual Computer pp. 1–15 (2024)

    2024

  25. [33]

    IEEE Computational Intelligence Magazine 12(4), 73–87 (2017)

    Tian, Y., Cheng, R., Zhang, X., Jin, Y.: Platemo: A matlab platform for evolutionary multi-objective optimization [edu- cational forum]. IEEE Computational Intelligence Magazine 12(4), 73–87 (2017). doi:10.1109/MCI.2017.2742868

    2017

  26. [34]

    IEEE Transactions on Systems, Man, and Cybernetics: Systems 51(9), 5880–5894 (2021)

    Tian, Y., He, C., Cheng, R., Zhang, X.: A multistage evolu- tionary algorithm for better diversity preservation in mul- tiobjective optimization. IEEE Transactions on Systems, Man, and Cybernetics: Systems 51(9), 5880–5894 (2021). doi:10.1109/TSMC.2019.2956288

    2021

  27. [35]

    International Journal of Science and Mathematics Education 12(1), 25–47 (2014)

    Wadouh, J., Liu, N., Sandmann, A., Neuhaus, B.J.: The Ef- fect Of Knowledge Linking Levels In Biology Lessons Upon Students’ Knowledge Structure. International Journal of Science and Mathematics Education 12(1), 25–47 (2014). doi:10.1007/s10763-012-9390-8

    2014 doi

  28. [36]

    The Visual Computer pp

    Wang, J., Wang, J., Li, H., Huang, X., Xia, J., Li, Z., Wu, W., Sheng, B.: Temporal goal-aware transformer assisted visual reinforcement learning for virtual table tennis agent. The Visual Computer pp. 1–15 (2025)

    2025

  29. [37]

    Frontiers in Psy- chology 12 (2021)

    Wang, W., Tu, Y., Song, L., Zheng, J., Wang, T.: An Adap- tive Design for Item Parameter Online Estimation and Q- Matrix Online Calibration in CD-CAT. Frontiers in Psy- chology 12 (2021). doi:10.3389/fpsyg.2021.710497

    2021

  30. [38]

    Wei, Y., Zhou, Y., Jiang, Y.H., Jiang, B.: Enhancing Ex- plainability of Knowledge Learning Paths: Causal Knowledge Networks. In: Joint Proceedings of the Human-Centric eX- plainable AI in Education and the Leveraging Large Lan- guage Models for Next Generation Educational Tec...

    2024

  31. [39]

    Educa- tional technology research and development 72(5), 2499– 2517 (2024)

    Weidlich, J., Hicks, B., Drachsler, H.: Causal reasoning with causal graphs in educational technology research. Educa- tional technology research and development 72(5), 2499– 2517 (2024). doi:10.1007/s11423-023-10241-0

    2024 doi

  32. [40]

    In: Proceedings of the 30th ACM SIGKDD Confer- ence on Knowledge Discovery and Data Mining, KDD ’24, pp

    Wen, Q., Liang, J., Sierra, C., Luckin, R., Tong, R., Liu, Z., Cui, P., Tang, J.: AI for education (AI4edu): Advanc- ing personalized education with LLM and adaptive learn- ing. In: Proceedings of the 30th ACM SIGKDD Confer- ence on Knowledge Discovery and Data Mining, KDD ’24...

    2024

  33. [41]

    World Journal on Educational Technology (2020)

    Yildiz, E.P., C ¸ engel, M., Alkan, A.: Current trends in ed- ucation technologies research worldwide: Meta-analysis of studies between 2015-2020. World Journal on Educational Technology (2020). URL http://acikerisim.subu.edu.tr/ xmlui/handle/20.500.14002/486

    2020

  34. [42]

    Analytical Biochemistry 656, 114,878 (2022)

    Yin, Y.H., Shen, L.C., Jiang, Y.H., Gao, S., Song, J., Yu, D.J.: Improving the prediction of dna-protein binding by integrating multi-scale dense convolutional network with fault-tolerant coding. Analytical Biochemistry 656, 114,878 (2022). doi:doi.org/10.1016/j.ab.2022.114878

    2022

  35. [43]

    URL https://lfs.aminer.cn/misc/moocdata/data/ mooccube2/prerequisites/math.json

    Yu, J., Wang, Y., Zhong, Q., Luo, G., Mao, Y., Sun, K., Feng, W., Xu, W., Cao, S., Zeng, K.: Mooccubex (2021). URL https://lfs.aminer.cn/misc/moocdata/data/ mooccube2/prerequisites/math.json

    2021

  36. [44]

    In: Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pp

    Yu, J., Wang, Y., Zhong, Q., Luo, G., Mao, Y., Sun, K., Feng, W., Xu, W., Cao, S., Zeng, K.: MOOCCubeX: A large knowledge-centered repository for adaptive learning in MOOCs. In: Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pp. 464...

    2021

  37. [45]

    The Visual Computer 41(2), 1097–1108 (2025)

    Zhou, Y., Chen, X., Li, T., Lin, S., Sheng, B., Liu, R., Dai, R.: Gamnet: a gated attention mechanism network for grad- ing myopic traction maculopathy in oct images. The Visual Computer 41(2), 1097–1108 (2025)

    2025

  38. [46]

    Zhuang, X., Wu, H., Shen, X., Yu, P., Yi, G., Chen, X., Hu, T., Chen, Y., Ren, Y., Zhang, Y., Song, Y., Liu, B., Lan, M.: TOREE: Evaluating Topic Relevance of Student Essays for Chinese Primary and Middle School Education. In: L.W. Ku, A. Martins, V. Srikumar (eds.) Findings o...

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.