Concept-Aware Latent and Explicit Knowledge Integration for Enhanced Cognitive Diagnosis

Huaxiang Zhang; Jiande Sun; Jing Li; Yawen Chen

REVIEW 4 major objections 4 minor 31 references

Concept-Aware Latent and Explicit Knowledge Integration for Enhanced Cognitive Diagnosis

T0 review · 4 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read CLEKI-CD claims that adding an attention-generated latent Q-matrix on top of the expert Q-matrix keeps cognitive diagnosis accurate even when the expert Q-matrix is sparse and knowledge concepts are fine-grained.

desk verdict The empirical gains look real, but the equations collapse the concept dimension—as written, the model cannot perform the per-concept diagnosis it claims. read the letter →

arxiv 2502.02104 v1 pith:QHILHODL submitted 2025-02-04 cs.LG

Yawen Chen , Jiande Sun , Jing Li , Huaxiang Zhang This is my paper · ORCID

classification cs.LG

keywords cognitivediagnosisQ-matrixknowledgecoveragegraphattentionnetworkconceptdependencymultidimensionalembeddingstudentmodelingintelligenttutoring

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

The paper tries to establish that cognitive diagnosis improves when student mastery and exercise difficulty are represented as multidimensional concept-aware vectors rather than scalars, and when the expert-defined Q-matrix is supplemented by a latent Q-matrix learned from relationships among knowledge concepts. The latent Q-matrix is built by graph attention over an expert-labeled concept dependency map, using cosine similarity and a mask to select the most relevant hidden concepts per explicit concept. The combined diagnosis layer weights predictions from both Q-matrices, and the full model reports the best ACC, AUC, and RMSE on ASSIST and Junyi across all train-test splits, with the gain clearest when practice logs are sparse. A sympathetic reader would care because the method addresses a known practical failure: as knowledge concepts become fine-grained, manually labeled Q-matrices become incomplete, and the model can still infer mastery of unseen concepts.

What carries the argument

The carrying mechanism is the latent Q-matrix $\tilde{Q} = Q\tilde{S}$, produced by a Graph Attention Network that aggregates knowledge-concept embeddings over an asymmetric adjacency matrix of expert-labeled dependency relations. The mask matrix $M$ (Eq. 7) keeps aggregation local, the top-$P$ selection with softmax (Eq. 8) keeps only the strongest latent relations, and the combined diagnosis layer (Eqs. 10–12) merges explicit and latent evidence with a balance coefficient $\epsilon$. All weights that map embeddings to proficiency and difficulty are constrained positive to preserve monotonicity.

What would settle it

Keep all components fixed but replace the concept dependency map with a randomly permuted graph or an empty graph; if ACC and AUC on ASSIST and Junyi do not drop materially, the reported advantage cannot be attributed to the latent Q-matrix mechanism.

Watch

Extended reading notes

Core claim

The central claim, stated in the paper's own terms, is that the binary expert Q-matrix is correct but incomplete, and that a latent Q-matrix $\tilde{Q} = Q\tilde{S}$ can fill the gap. The paper constructs $\tilde{Q}$ by attention-based aggregation on a graph whose asymmetric adjacency distinguishes prerequisite from similarity relations, computing cosine similarities between concept embeddings, masking unrelated pairs, retaining the top-$P$ similarities per row, and softmax-normalizing them. The latent matrix is then combined with the explicit matrix through a weighted diagnostic layer $y = \epsilon u + (1-\epsilon)\tilde{u}$, where each $u$ is a monotone interaction of $L^2$-normed student proficiency and exercise difficulty vectors. On ASSIST and Junyi, this combination outperforms seven baselines under 80/20, 70/30, and 60/40 splits, and degrades less than baselines as knowledge coverage shrinks.

Load-bearing premise

An expert-labeled concept dependency map exists and is correct for each dataset; the paper does not say who built it or how it was validated, and without it the attention aggregation has no meaningful neighborhood to draw latent concepts from.

Editorial extensions

If this is right

Cognitive diagnosis models can work with sparser expert Q-matrices, mitigating the knowledge coverage problem when fine-grained concepts are introduced.
The model can infer mastery of unpracticed concepts through concept correlations, so student diagnosis reports can include concepts the student never directly exercised.
The performance advantage grows as training data and coverage decrease, which makes the approach useful for low-resource tutoring systems with short response logs.
The diagnosis of a concept can be explained through the top latent concepts associated with it, supporting interpretable feedback in educational practice.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the claim transfers, an automatic pipeline that mines prerequisite and similarity relations from logs could replace expert maps and make the latent Q-matrix construction applicable to domains without curriculum taxonomies.
The top-$P$ latent concepts and their cosine weights are produced per dataset; inspecting whether those weights align with human judgments in a real curriculum would test the interpretability story beyond the single case study.
The balance coefficient $\epsilon$ is fixed globally in the paper; learning it per student or per exercise could reveal whether explicit and latent knowledge should be weighted differently for low- versus high-mastery students.
The same latent-Q construction could be dropped into knowledge-tracing or dynamic diagnosis models where concept coverage changes over time.

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.

Desk Editor's Note

read the letter

Short version: the paper reports consistent small gains, but the formal core of the model does not do what it says. Equations 10 and 11 collapse the concept dimension because the L2 norms of the K×D matrices are scalars independent of k, so the Q-matrix (explicit or latent) enters only through its row sum. That means the per-concept diagnosis and interpretability claims are not supported by the math.

What's actually new: the combination of multidimensional concept-wise embeddings, a GAT over a dependency graph to produce a latent Q-matrix, and a weighted fusion of the two diagnostic paths. That package isn't in the cited literature. The experiments are consistent across two datasets and three splits, with gains of about 1–2 ACC/AUC over RCD and ORCDF. The robustness experiment under reduced coverage is a nice empirical touch.

Soft spots: the concept-dependency map is expert-labeled and never described for ASSIST or Junyi, so the main input to the GAT is unverifiable. There are no error bars or significance tests. But those are secondary. The primary problem is Eq. 10–11. If we read h^s_i and h^diff_j as matrices, their L2 norms are scalars, so the sum over k multiplies a k-independent scalar by q^k_j, and dividing by c_j cancels the Q-row sum. The final prediction is a logistic function of a scalar per student, a scalar per exercise, and a per-exercise weight from the Q-matrix row sum. The concept index disappears entirely. This is not a minor typo: the paper's per-concept bar charts, case study, and 'rationality analysis' cannot be derived from the model as written. A row-wise norm (e.g., ||h_{i,k}||_2 over D) would restore concept sensitivity, and the authors should correct the notation and release code.

If the intended row-wise norm is what they meant, the model is a plausible incremental contribution, and the empirical gains might survive. As it stands, the formalization fails to support the central claims, so the paper needs major revision before it is publishable.

Recommendation: send it to peer review with the expectation of heavy revision; the issue is fixable and the empirical work is worth checking.

Referee Report

4 major / 4 minor

Summary. The manuscript proposes CLEKI-CD, a cognitive diagnosis model that represents student proficiency and exercise difficulty as concept-aware multidimensional vectors, generates a latent Q-matrix via graph attention over a concept dependency map, and combines explicit and latent knowledge through a weighted diagnostic layer. The authors report experiments on ASSIST and Junyi across three train/test splits, claiming consistent improvements in ACC, AUC, and RMSE over several baselines, along with ablations, case studies, embedding visualizations, and hyperparameter analyses supporting interpretability and robustness under sparse coverage.

Significance. If the proposed mechanism worked as described, the paper would offer a modest but consistent improvement over existing neural cognitive diagnosis models, particularly under sparse interaction, and would contribute a way to supplement incomplete expert Q-matrices with latent concept relations. The paper also attempts interpretability through case studies and visualizations, which is valuable for the educational application context. However, the central mathematical formulation in Eqs. (10)-(11) collapses the concept dimension, so the claimed concept-aware mechanism is not supported as written. The load-bearing dependency on an expert-labeled concept map is also not documented for the datasets used. The empirical claims would need a corrected derivation, error bars, and a reproducible implementation to be assessed fairly.

major comments (4)

[Section 3.3, Eqs. (10)-(11)] The norms ||h_i^s||_2 and ||h_j^diff||_2 are taken over the full K×D matrices and therefore have no concept index k. In Eq. (10), the sum over k of q_j^k times a k-independent scalar equals c_j times that scalar, so after division by c_j the explicit diagnosis reduces to σ(||h_i^s||_2 − ||h_j^diff||_2). The same collapse occurs in Eq. (11) for the latent Q-matrix. Consequently, the model's predictions depend only on the row sum of the explicit and latent Q-matrices, not on which concepts an exercise involves. This directly contradicts the paper's central claims that CLEKI-CD diagnoses each knowledge concept (Sec. 4.4, 4.7) and that the latent Q-matrix supplements sparse per-concept coverage. The authors must correct the notation to use row-wise norms, e.g., ||h_i^s[k,:]||_2, and re-derive the diagnostic equations, or substantially weaken the claims about concept-aware diagnosis. As written, the reported gains could be reproduced by a baseline with one global student scalar, one global difficulty scalar, and a per-exercise learned scale, none of which require a concept dependency map.
[Section 2.2 and Eq. (7)] The expert-labeled concept dependency map is load-bearing: it defines the graph attention neighborhood in Eq. (4) and the mask matrix M in Eq. (7). However, the paper never states how the prerequisite and similarity relations were obtained for ASSIST and Junyi, how many edges they contain, who labeled them, or whether they were validated. For Junyi, Table 1 reports K=734 knowledge concepts with M=734 exercises and one concept per exercise, so the dependency map over 734 concepts is particularly nontrivial. Without this information, the latent Q-matrix generation is not reproducible, and the reported gains cannot be separated from the particular choice of concept map. The authors should describe the construction, release the map, and ideally ablate its influence by comparing with alternative maps or learned adjacency.
[Table 2] All empirical results are reported as single runs without standard deviations or significance tests. The improvements over the best baseline are small in several settings (e.g., ACC 74.35 vs. 73.09 and AUC 77.89 vs. 76.40 on ASSIST 80/20; RMSE 41.70 vs. 42.25). Given this, the claim that CLEKI-CD 'outperforms the state-of-the-art models' is not statistically supported. The authors should report means and standard deviations over multiple seeds and perform paired significance tests (e.g., t-test or Wilcoxon) for the key comparisons.
[Sections 4.4 and 4.7] The case studies claim per-concept diagnostic results, including proficiency and difficulty for individual concepts and for 'latent knowledge concepts most relevant to it.' However, because Eqs. (10)-(11) produce only exercise-level scalars, the paper does not derive these per-concept values from the model equations. Figure 3 and Figure 6 therefore appear to be post-hoc visualizations that are not clearly generated by the model. The authors should specify exactly how the per-concept bars and relevance scores are computed, or remove these interpretability claims unless they follow from a corrected formulation.

minor comments (4)

[Section 3.2, Eq. (5)] The notation in Eq. (5) uses h in the superscript of x (e.g., Wh xh_i and Wh xh_j), which is inconsistent with the earlier definition of xk_i. Please use a single consistent notation for the concept embeddings.
[Section 1] There is a typo: 'serval studies' should be 'several studies.'
[Section 4.6] Figure 5 is described in the text as showing performance under varying knowledge coverage rates via gradually reducing students' response logs, but the caption says 'varying training data proportions.' Please clarify whether the x-axis denotes training data proportion or coverage rate, as these are different manipulations.
[Section 4.8] The hyperparameter analysis reports optimal top-k=8 and an optimal epsilon, but the grid range and step size for these hyperparameters are not given. Please include the search range for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the latent Q-matrix is learned from response logs via GAT, the loss is on training logs, and evaluation is on held-out responses.

full rationale

CLEKI-CD's derivation chain is not circular. The explicit Q-matrix is an external, expert-labeled input; the concept dependency map is a stated assumption (Section 2.2); the latent Q-matrix is computed from that input through GAT aggregation and cosine similarity (Eqs. 4-9); and the combined diagnosis layer (Eqs. 10-12) is trained with cross-entropy (Eq. 13) on response logs. Predictive accuracy is measured on held-out response logs (Section 4.1), so no training target is reused as a prediction and no fitted parameter is renamed as a discovery. The paper also does not rest its central claim on self-citations: the cited GAT, IRT/MIRT, and Q-matrix works are external baselines or building blocks. The main caveat—that the expert-labeled concept dependency map and mask M (Eq. 7) are not described or validated for ASSIST/Junyi—is an assumption and reproducibility concern, not circularity. Separately, the norms in Eqs. 10-11 are k-independent, so the concept summation cancels (Σ q^k_j = c_j, and each row of S̃ sums to 1), making the per-concept diagnosis collapse to a scalar σ(||h^s_i||_2 − ||h^diff_j||_2); this is a serious internal-consistency/correctness flaw, but it does not make the held-out comparison circular.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim depends on two tuned hyperparameters and on domain assumptions about the correctness of expert maps. The latent Q-matrix is constructed from learned embeddings, making it a fitted component rather than an independent discovery.

free parameters (2)

top-P (top-k) in latent Q similarity selection = 8 (optimal on ASSIST)
Selected by hyperparameter analysis in Section 4.8; changes the number of latent concepts aggregated into the latent Q-matrix.
balance coefficient epsilon = not reported explicitly, tuned
Weights explicit versus latent knowledge in Eq. (12); tuned per Section 4.8.

assumptions (4)

domain assumption The expert-labeled Q-matrix is correct but may be incomplete.
Stated in Section 3.2; if the Q-matrix has errors, the latent generation compounds them.
domain assumption An expert-labeled concept dependency map (prerequisite and similarity relations) is available and correct for ASSIST and Junyi.
The GAT aggregation and mask rely on this map (Sections 2.2, 3.2), but its source is never described.
domain assumption Monotonicity of correct-answer probability in student mastery is preserved by the norm-difference interaction and positive weight matrices.
The paper asserts monotonicity (Section 3.1) but does not prove it for the combined prediction in Eq. (12).
ad hoc to paper Cosine similarity of GAT-aggregated knowledge embeddings captures latent concept relations.
This is the core mechanism of the latent Q-matrix; no independent validation is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Concept-Aware Latent and Explicit Knowledge Integration for Enhanced Cognitive Diagnosis." pith.science (2026). https://pith.science/paper/QHILHODL

@misc{pith2026250202104,
  author       = {Pith},
  title        = {Pith review of: Concept-Aware Latent and Explicit Knowledge Integration for Enhanced Cognitive Diagnosis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QHILHODL}},
  note         = {Machine review of arXiv:2502.02104}
}

read the original abstract

Cognitive diagnosis can infer the students' mastery of specific knowledge concepts based on historical response logs. However, the existing cognitive diagnostic models (CDMs) represent students' proficiency via a unidimensional perspective, which can't assess the students' mastery on each knowledge concept comprehensively. Moreover, the Q-matrix binarizes the relationship between exercises and knowledge concepts, and it can't represent the latent relationship between exercises and knowledge concepts. Especially, when the granularity of knowledge attributes refines increasingly, the Q-matrix becomes incomplete correspondingly and the sparse binary representation (0/1) fails to capture the intricate relationships among knowledge concepts. To address these issues, we propose a Concept-aware Latent and Explicit Knowledge Integration model for cognitive diagnosis (CLEKI-CD). Specifically, a multidimensional vector is constructed according to the students' mastery and exercise difficulty for each knowledge concept from multiple perspectives, which enhances the representation capabilities of the model. Moreover, a latent Q-matrix is generated by our proposed attention-based knowledge aggregation method, and it can uncover the coverage degree of exercises over latent knowledge. The latent Q-matrix can supplement the sparse explicit Q-matrix with the inherent relationships among knowledge concepts, and mitigate the knowledge coverage problem. Furthermore, we employ a combined cognitive diagnosis layer to integrate both latent and explicit knowledge, further enhancing cognitive diagnosis performance. Extensive experiments on real-world datasets demonstrate that CLEKI-CD outperforms the state-of-the-art models. The proposed CLEKI-CD is promising in practical applications in the field of intelligent education, as it exhibits good interpretability with diagnostic results.

Figures

Figures reproduced from arXiv: 2502.02104 by the authors.

**Figure 2.** The overall structure of proposed CLEKI-CD. The Embedding Modules capture richer and more expressive features of student [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 4.** The t-SNE visualization of (a) students’ mastery and (b) [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

Figures from the paper (3 more)

**Figure 3.** Figure 3: Proficiency and difficulty on (a) knowledge concepts A-E, [PITH_FULL_IMAGE:figures/full_fig_p006_3.png]

**Figure 7.** Figure 7: Hyperparameter analysis of (a) top-k and (b) [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

**Figure 6.** Figure 6: The results of student diagnosis report. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 28 canonical work pages

[1]

Multidimensional item response theory models

[Ackerman, 2014] Terry A Ackerman. Multidimensional item response theory models. Wiley StatsRef: Statistics Reference Online,

work page 2014
[10]

The attribute hierarchy method for cognitive assessment: A variation on tatsuoka’s rule- space approach

[Leighton et al., 2004] Jacqueline P Leighton, Mark J Gierl, and Stephen M Hunka. The attribute hierarchy method for cognitive assessment: A variation on tatsuoka’s rule- space approach. Journal of educational measurement , 41(3):205–237,

work page 2004
[11]

Fuzzy cognitive diagnosis for modelling examinee performance

[Liu et al., 2018] Qi Liu, Runze Wu, Enhong Chen, Guan- dong Xu, Yu Su, Zhigang Chen, and Guoping Hu. Fuzzy cognitive diagnosis for modelling examinee performance. ACM Transactions on Intelligent Systems and Technology (TIST), 9(4):1–26,

work page 2018
[13]

Inductive cognitive diagnosis for fast student learning in web-based intelligent education systems

[Liu et al., 2024] Shuo Liu, Junhao Shen, Hong Qian, and Aimin Zhou. Inductive cognitive diagnosis for fast student learning in web-based intelligent education systems. In Proceedings of the ACM on Web Conference 2024 , pages 4260–4271,

work page 2024
[15]

Design and evaluation of trustworthy knowl- edge tracing model for intelligent tutoring system

[Lu et al., 2024] Yu Lu, Deliang Wang, Penghe Chen, and Zhi Zhang. Design and evaluation of trustworthy knowl- edge tracing model for intelligent tutoring system. IEEE Transactions on Learning Technologies,

work page 2024
[16]

Knowledge-sensed cognitive diagnosis for intelligent ed- ucation platforms

[Ma et al., 2022] Haiping Ma, Manwei Li, Le Wu, Haifeng Zhang, Yunbo Cao, Xingyi Zhang, and Xuemin Zhao. Knowledge-sensed cognitive diagnosis for intelligent ed- ucation platforms. In Proceedings of the 31st ACM inter- national conference on information & knowledge manage- ment, pages 1451–1460,

work page 2022
[17]

Dgcd: an adaptive denoising gnn for group-level cog- nitive diagnosis

[Ma et al., 2024] Haiping Ma, Siyu Song, Chuan Qin, Xi- aoshan Yu, Limiao Zhang, Xingyi Zhang, and Hengshu Zhu. Dgcd: an adaptive denoising gnn for group-level cog- nitive diagnosis. In The International Joint Conference on Artificial Intelligence,

work page 2024
[18]

Area under the curve as a mea- sure of discounting

[Myerson et al., 2001] Joel Myerson, Leonard Green, and Missaka Warusawitharana. Area under the curve as a mea- sure of discounting. Journal of the experimental analysis of behavior, 76(2):235–243,

work page 2001

Show all 31 references

[20]

Modeling relational data with graph con- volutional networks

[Schlichtkrull et al., 2017] Michael Schlichtkrull, Thomas N Kipf, Peter Bloem, Rianne van den Berg, Ivan Titov, and Max Welling. Modeling relational data with graph con- volutional networks. arXiv preprint arXiv:1703.06103 ,

2017 arXiv
[22]

A deep cross-modal neural cognitive diagnosis framework for modeling student performance

[Song et al., 2023] Lingyun Song, Mengting He, Xuequn Shang, Chen Yang, Jun Liu, Mengzhen Yu, and Yu Lu. A deep cross-modal neural cognitive diagnosis framework for modeling student performance. Expert Systems with Applications, 230:120675,

2023
[23]

Item response ranking for cognitive diagnosis

[Tong et al., 2021] Shiwei Tong, Qi Liu, Runlong Yu, Wei Huang, Zhenya Huang, Zachary A Pardos, and Weijie Jiang. Item response ranking for cognitive diagnosis. In The International Joint Conference on Artificial Intelli- gence,

2021
[24]

Graph attention networks

[Veliˇckovi´c et al., 2018] Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li `o, and Yoshua Bengio. Graph attention networks. In Interna- tional Conference on Learning Representations,

2018
[25]

Neural cognitive diagnosis for intelligent educa- tion systems

[Wang et al., 2020] Fei Wang, Qi Liu, Enhong Chen, Zhenya Huang, Yuying Chen, Yu Yin, Zai Huang, and Shijin Wang. Neural cognitive diagnosis for intelligent educa- tion systems. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 6153–6161,

2020
[26]

Neuralcd: a gen- eral framework for cognitive diagnosis

[Wang et al., 2022] Fei Wang, Qi Liu, Enhong Chen, Zhenya Huang, Yu Yin, Shijin Wang, and Yu Su. Neuralcd: a gen- eral framework for cognitive diagnosis. IEEE Transac- tions on Knowledge and Data Engineering , 35(8):8312– 8327,

2022
[27]

Self-supervised graph learning for long-tailed cognitive diagnosis

[Wang et al., 2023] Shanshan Wang, Zhen Zeng, Xun Yang, and Xingyi Zhang. Self-supervised graph learning for long-tailed cognitive diagnosis. In Proceedings of the AAAI conference on artificial intelligence , volume 37, pages 110–118,

2023
[28]

A survey of models for cognitive diagnosis: New developments and future di- rections

[Wang et al., 2024] Fei Wang, Weibo Gao, Qi Liu, Jiatong Li, Guanhao Zhao, Zheng Zhang, Zhenya Huang, Mengx- iao Zhu, Shijin Wang, Wei Tong, et al. A survey of models for cognitive diagnosis: New developments and future di- rections. arXiv preprint arXiv:2407.05458,

2024 arXiv
[29]

Advantages of the mean absolute error (mae) over the root mean square error (rmse) in assessing aver- age model performance

[Willmott and Matsuura, 2005] Cort J Willmott and Kenji Matsuura. Advantages of the mean absolute error (mae) over the root mean square error (rmse) in assessing aver- age model performance. Climate research, 30(1):79–82,

2005
[2001]

Orcdf: An oversmoothing-resistant cognitive diagnosis framework for student learning in online education systems

[Qian et al., 2024] Hong Qian, Shuo Liu, Mingjia Li, Bing- dong Li, Zhi Liu, and Aimin Zhou. Orcdf: An oversmoothing-resistant cognitive diagnosis framework for student learning in online education systems. In Pro- ceedings of the 30th ACM SIGKDD Conference on Knowl- edge Disc...

2024
[2005]

Hierarchical attention networks for document classification

[Yang et al., 2016] Zichao Yang, Diyi Yang, Chris Dyer, Xi- aodong He, Alex Smola, and Eduard Hovy. Hierarchical attention networks for document classification. In Pro- ceedings of the 2016 conference of the North American chapter of the association for computational linguisti...

2016
[2009]

Rcd: Relation map driven cognitive diagnosis for intelligent education systems

[Gao et al., 2021] Weibo Gao, Qi Liu, Zhenya Huang, Yu Yin, Haoyang Bi, Mu-Chun Wang, Jianhui Ma, Shi- jin Wang, and Yu Su. Rcd: Relation map driven cognitive diagnosis for intelligent education systems. In Proceed- ings of the 44th international ACM SIGIR conference on resear...

2021
[2014]

Cog- nitive diagnostic model made more practical by genetic al- gorithm

[Bu et al., 2022] Chenyang Bu, Fei Liu, Zhiyong Cao, Lei Li, Yuhong Zhang, Xuegang Hu, and Wenjian Luo. Cog- nitive diagnostic model made more practical by genetic al- gorithm. IEEE Transactions on Emerging Topics in Com- putational Intelligence, 7(2):447–461,

2022
[2015]

Dirt: Deep learning enhanced item response theory for cognitive diagnosis

[Cheng et al., 2019] Song Cheng, Qi Liu, Enhong Chen, Zai Huang, Zhenya Huang, Yiying Chen, Haiping Ma, and Guoping Hu. Dirt: Deep learning enhanced item response theory for cognitive diagnosis. In Proceedings of the 28th ACM international conference on information and knowl- ...

2019
[2016]

Item response theory

[Yen and Fitzpatrick, 2006] Wendy M Yen and Anne R Fitz- patrick. Item response theory. Educational measurement, 4:111–153, 2006

2006
[2017]

Improving cognitive diagnosis models with adaptive relational graph neural networks

[Shao et al., 2024] Pengyang Shao, Chen Gao, Lei Chen, Yonghui Yang, Kun Zhang, and Meng Wang. Improving cognitive diagnosis models with adaptive relational graph neural networks. arXiv preprint arXiv:2403.05559,

2024 arXiv
[2018]

Qccdm: A q-augmented causal cognitive di- agnosis model for student learning

[Liu et al., 2023] Shuo Liu, Hong Qian, Mingjia Li, and Aimin Zhou. Qccdm: A q-augmented causal cognitive di- agnosis model for student learning. In ECAI 2023, pages 1536–1543. IOS Press,

2023
[2019]

Dina model and parameter estimation: A didactic

[De La Torre, 2009] Jimmy De La Torre. Dina model and parameter estimation: A didactic. Journal of educational and behavioral statistics, 34(1):115–130,

2009
[2020]

Group-level cognitive diagnosis: A multi-task learning perspective

[Huang et al., 2021] Jie Huang, Qi Liu, Fei Wang, Zhenya Huang, Songtao Fang, Runze Wu, Enhong Chen, Yu Su, and Shijin Wang. Group-level cognitive diagnosis: A multi-task learning perspective. In 2021 IEEE Interna- tional Conference on Data Mining (ICDM) , pages 210–

2021
[2021]

Leveraging transferable knowledge concept graph embedding for cold-start cognitive diagnosis

[Gao et al., 2023] Weibo Gao, Hao Wang, Qi Liu, Fei Wang, Xin Lin, Linan Yue, Zheng Zhang, Rui Lv, and Shijin Wang. Leveraging transferable knowledge concept graph embedding for cold-start cognitive diagnosis. In Proceed- ings of the 46th international ACM SIGIR conference on ...

2023
[2022]

Modeling exercise relationships in e-learning: A unified approach

[Chang et al., 2015] Haw-Shiuan Chang, Hwai-Jung Hsu, Kuan-Ta Chen, et al. Modeling exercise relationships in e-learning: A unified approach. In EDM, pages 532–535,

2015
[2023]

Learning or forgetting? a dynamic approach for tracking the knowledge proficiency of students

[Huang et al., 2020] Zhenya Huang, Qi Liu, Yuying Chen, Le Wu, Keli Xiao, Enhong Chen, Haiping Ma, and Guop- ing Hu. Learning or forgetting? a dynamic approach for tracking the knowledge proficiency of students. ACM Transactions on Information Systems (TOIS), 38(2):1–33,

2020
[2024]

Towards a new generation of cognitive diagnosis

[Liu, 2021] Qi Liu. Towards a new generation of cognitive diagnosis. In The International Joint Conference on Arti- ficial Intelligence,

2021

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

Multidimensional item response theory models

[Ackerman, 2014] Terry A Ackerman. Multidimensional item response theory models. Wiley StatsRef: Statistics Reference Online,

work page 2014

[2] [10]

The attribute hierarchy method for cognitive assessment: A variation on tatsuoka’s rule- space approach

[Leighton et al., 2004] Jacqueline P Leighton, Mark J Gierl, and Stephen M Hunka. The attribute hierarchy method for cognitive assessment: A variation on tatsuoka’s rule- space approach. Journal of educational measurement , 41(3):205–237,

work page 2004

[3] [11]

Fuzzy cognitive diagnosis for modelling examinee performance

[Liu et al., 2018] Qi Liu, Runze Wu, Enhong Chen, Guan- dong Xu, Yu Su, Zhigang Chen, and Guoping Hu. Fuzzy cognitive diagnosis for modelling examinee performance. ACM Transactions on Intelligent Systems and Technology (TIST), 9(4):1–26,

work page 2018

[4] [13]

Inductive cognitive diagnosis for fast student learning in web-based intelligent education systems

[Liu et al., 2024] Shuo Liu, Junhao Shen, Hong Qian, and Aimin Zhou. Inductive cognitive diagnosis for fast student learning in web-based intelligent education systems. In Proceedings of the ACM on Web Conference 2024 , pages 4260–4271,

work page 2024

[5] [15]

Design and evaluation of trustworthy knowl- edge tracing model for intelligent tutoring system

[Lu et al., 2024] Yu Lu, Deliang Wang, Penghe Chen, and Zhi Zhang. Design and evaluation of trustworthy knowl- edge tracing model for intelligent tutoring system. IEEE Transactions on Learning Technologies,

work page 2024

[6] [16]

Knowledge-sensed cognitive diagnosis for intelligent ed- ucation platforms

[Ma et al., 2022] Haiping Ma, Manwei Li, Le Wu, Haifeng Zhang, Yunbo Cao, Xingyi Zhang, and Xuemin Zhao. Knowledge-sensed cognitive diagnosis for intelligent ed- ucation platforms. In Proceedings of the 31st ACM inter- national conference on information & knowledge manage- ment, pages 1451–1460,

work page 2022

[7] [17]

Dgcd: an adaptive denoising gnn for group-level cog- nitive diagnosis

[Ma et al., 2024] Haiping Ma, Siyu Song, Chuan Qin, Xi- aoshan Yu, Limiao Zhang, Xingyi Zhang, and Hengshu Zhu. Dgcd: an adaptive denoising gnn for group-level cog- nitive diagnosis. In The International Joint Conference on Artificial Intelligence,

work page 2024

[8] [18]

Area under the curve as a mea- sure of discounting

[Myerson et al., 2001] Joel Myerson, Leonard Green, and Missaka Warusawitharana. Area under the curve as a mea- sure of discounting. Journal of the experimental analysis of behavior, 76(2):235–243,

work page 2001