Pith. sign in

REVIEW 3 major objections 4 minor 60 references

Combining Log Data and Collaborative Dialogue Features to Predict Project Quality in Middle School AI Education

T0 review · 3 major / 4 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 This paper finds that in collaborative chatbot building, the best data source for predicting project quality depends on the outcome: logs estimate how much students produce, dialogue estimates how content-rich their phrases are.

desk verdict A competent workshop paper whose central fusion claim is partly an artifact of feature-label leakage; the authors disclose it and ablate it, but the main table still reports the leaky numbers. read the letter →

arxiv 2506.11326 v1 pith:TNNT267A submitted 2025-06-12 cs.HC

Conrad Borchers , Xiaoyi Tian , Kristy Elizabeth Boyer , Maya Israel This is my paper · ORCID
classification cs.HC
keywords multimodallearninganalyticsproject-basedcollaborativechatbotdevelopmentmiddleschoolAIeducationlogdatadialogueanalysisprojectqualityprediction
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 in project-based AI learning, the best predictor of project quality depends on which quality you are predicting. In a study of 94 middle school students building chatbots in pairs, the authors predict three properties of each chatbot intent: how many training phrases the students wrote (productivity), how content-rich those phrases are (lexical density), and how varied their wording is (lexical variation). Logs of system interactions best predict productivity, while transcribed dialogue best predicts content richness; both modalities predict lexical variation only modestly. Combining the two modalities improves productivity and lexical variation predictions but not content richness. The implication is that multimodal fusion is only useful for some learning outcomes, so analytics systems should match the data modality to the target construct.

What carries the argument

The central unit is the intent-working segment: the time window from when a pair begins working on one chatbot intent until they submit it for training, with dialogue transcripts and system logs aligned to within one second. Log features summarize timing statistics and event n-grams over 31 interaction event types; dialogue features are 768-dimensional sentence embeddings of the transcript for that segment. These features are fed into a feedforward neural network whose holdout AUC is compared across log-only, dialogue-only, and combined inputs, with early fusion (concatenation) used for the combined model. The three outcome measures are computed directly from the students' submitted training phrases.

What would settle it

Re-run the same prediction pipeline with leave-one-pair-out folds, holding out all intents from entire pairs; if the log-only AUC for training phrase count drops markedly while the dialogue-only AUC stays similar, the reported advantage of logs comes from pair leakage rather than from genuine predictive signal.

Watch

Extended reading notes

Core claim

For 47 pairs of middle school students collaborating on chatbot projects, the paper reports that a feedforward neural network trained on system interaction logs reaches a holdout AUC of 0.81 for predicting training phrase count, whereas dialogue-derived embeddings alone reach 0.60; for lexical density, the dialogue-only model reaches 0.66 while logs alone are at chance (0.51). Both single modalities are modest for lexical variation (logs 0.60, dialogue 0.53). Concatenating log and dialogue features improves training phrase count to 0.83 and lexical variation slightly to 0.61, but it lowers lexical density to 0.57. The paper concludes that the two data sources carry complementary signals and that the value of fusion is outcome-dependent, which matters for designing classroom monitoring tools for open-ended project work.

Load-bearing premise

The analysis assumes that the five-fold student-level cross-validation keeps each collaborating pair together in one fold, so that the same pair's data never appears in both training and test sets.

Editorial extensions

If this is right

  • A detector built for one project-quality outcome should not be assumed to work for another: log-based monitoring fits productivity, dialogue-based monitoring fits content richness.
  • Multimodal fusion should be applied selectively, because adding log features to dialogue features can reduce accuracy for content richness.
  • An automated pipeline that aligns transcripts and logs could support offline teacher dashboards that flag intents where students are active but their training phrases are lexically thin.
  • Improving prediction of lexical variation will likely require additional features beyond current logs and whole-segment dialogue embeddings, such as turn-taking dynamics.

Reading between the lines

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

  • The paper's ablation, which removes log features that directly count phrase additions and deletions, cuts the combined-model training-phrase-count AUC from 0.83 to 0.68, so much of the fusion gain may come from directly counting phrase additions rather than from deeper behavioral insight.
  • Because the paper does not specify whether both members of each pair stay in the same cross-validation fold, the reported AUCs may be optimistic; a pair-level split is the natural robustness check.
  • A testable extension is to separate driver and navigator dialogue turns, since pair-programming roles likely differ in how their talk predicts the quality of the submitted phrases.
  • The outcome-dependent pattern suggests a practical design principle for classroom analytics: choose the data stream that matches the construct, and use fusion only when the secondary stream adds orthogonal signal.
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

3 major / 4 minor

Summary. This paper predicts three intent-level project-quality outcomes (training phrase count as productivity, lexical density as content richness, and lexical variation) using dialogue transcript embeddings and system log features from 94 middle school students working in 47 pairs over 121 chatbot-development sessions. The authors compare log-only, dialogue-only, and early-fusion models, reporting that log data best predicts training phrase count, dialogue data best predicts lexical density, and fusion improves productivity and lexical variation but not content richness. A footnote discloses a feature-label leakage concern for training phrase count and provides an ablation that substantially changes the fusion result.

Significance. The topic is timely for multimodal learning analytics in K-12 AI education, and the authors should be credited for including a validation of the outcome measures (learning curves and correlations with expert ratings), for releasing code in a digital appendix, and for disclosing the leakage issue in a footnote with an ablation. If the central claims survive revision, the paper would provide a useful comparison of log and dialogue modalities for assessing process-level project quality. However, the central RQ2 result for productivity is compromised by feature-label leakage, and the paper's own ablation shows that the fusion advantage disappears once the leaky features are removed; the main text and abstract currently overstate the finding.

major comments (3)
  1. [§4.1, Table 1, §5.2, footnote 3] The log feature set includes 2,124 event-based features built from 31 event types, which include 'add-training-phrase', while the outcome 'training phrase count' is literally the number of such phrases. The single-modality log model can therefore read off the outcome directly. The ablation in footnote 3 confirms this: removing features dependent on 'add-training-phrase', 'delete-phrase', and 'add-response' drops log-only AUC from 0.8053 to 0.6865 and combined AUC from 0.8301 to 0.6759. With ablated features, the combined model is worse than log-only (0.6759 < 0.6865), so the paper's conclusion that multimodal fusion improves productivity predictions no longer holds. The abstract, Section 5.2, and Table 1 should report the ablated values as the primary results, and the fusion claim should be revised or removed.
  2. [§4.2] The evaluation section reports 'five-fold student-level cross-validation' but does not state whether all sessions from the same collaborating pair are kept in one fold. Because the unit of analysis is intent submissions nested within pair sessions, splitting a pair across training and test folds would allow the model to see the same pair's dialogue and log behavior on both sides, which can inflate the reported AUCs. Please clarify the exact grouping variable (pair, session, or student) and, if pairs were split, rerun the evaluation with pair-level folds.
  3. [Table 1] Even setting the leakage issue aside, the claimed fusion improvement for training phrase count is not supported by the reported confidence intervals: the combined AUC interval [0.7732, 0.8822] overlaps the log-only interval [0.7470, 0.8604]. The paper should either provide a paired significance test across folds or soften the claim from 'improved' to 'directionally higher'.
minor comments (4)
  1. [§3.1 and §3.2] The paper reports 100 consented students, 97 with demographic information, but the dataset description uses 94 individuals in 47 pairs; please reconcile these numbers explicitly.
  2. [Footnote 3] The ablation is reported in a footnote after the results; it should be moved to the main text with details on the feature removal procedure and full results for all three outcomes, not only training phrase count.
  3. [§3.3] The 'ad-hoc decision' to remove segments longer than 12.5 minutes is justified only by a claimed lack of correlation with outcome measures; please provide a more direct analysis of exclusion bias, such as comparing model performance with and without the filter.
  4. [§4.1] Please report the final number of intent-level instances after segment removal and feature engineering, so readers can assess the overfitting risk of using 2,124 log features and 768-dimensional dialogue embeddings with a feedforward network.

Circularity Check

2 steps flagged · score 6.0 of 10

Log features for training-phrase count include add-training-phrase event counts—the outcome itself—so the headline productivity AUC and fusion gain are partly circular; the authors' footnote ablation confirms the fusion gain disappears.

  1. self definitional [Section 4.1 (Log Data Features); Section 3.4 (Outcome Measures); Footnote 3; Table 1]
    "Event-based features (2124 features) characterized interaction patterns through event type counts and bigram/trigram sequences of the 31 unique event types (e.g., [test-chatbot, create-new-intent, add-phrase]). Training Phrase Count measures the number of phrases input by students for training the chatbot to recognize an intent. As one attentive reviewer pointed out, counting ’add-training-phrase’ logs might straightforwardly reveal the number of training phrases."

    The log feature set includes event-type counts of 'add-phrase'/'add-training-phrase' events within each intent-working segment, while the predicted outcome is the number of training phrases submitted for that same intent. A model can therefore read off the outcome from a single feature, making the log-only AUC (0.8053) for training phrase count partially a self-fulfilling construction rather than an independent behavioral signal. The authors' ablation confirms this: removing features dependent on add-training-phrase, delete-phrase, and add-response drops log-only AUC to 0.6865, showing a large share of the predictive signal was the outcome count itself.

  2. fitted input called prediction [Section 5.2 (RQ2); Footnote 3; Table 1]
    "for training phrase count, fusion improved AUC from 0.8053 (log-only) to 0.8301. ... As one attentive reviewer pointed out, counting ’add-training-phrase’ logs might straightforwardly reveal the number of training phrases. We performed an ablation study, recomputing these accuracies without features dependent on ’add-training-phrase’, ’delete-phrase’, and ’add-response’. For completeness outcome, AUCs changed to 0.6865 (log only) and 0.6759 in the combined modalities, respectively."

    The reported fusion benefit for productivity (0.8301 vs 0.8053) is driven by the leaky add-training-phrase count that appears in both the log-only and combined feature sets. Under the authors' own ablation, the combined model (0.6759) is actually worse than log-only (0.6865), so the central RQ2 claim that multimodal fusion improved productivity predictions is not supported once the outcome-derived feature is removed. The abstract and Section 5.2 continue to report the unablated numbers as the main result, presenting a leakage artifact as a genuine cross-modal gain.

full rationale

The training-phrase-count outcome is partially circular because the log feature set directly includes counts of add-training-phrase events, which are the very quantity being predicted. The authors disclose this in footnote 3 and provide an ablation, which is good practice, but they do not revise the headline productivity and fusion claims; the ablation shows the fusion advantage reverses (combined 0.6759 < log-only 0.6865). The other two outcomes, lexical density and lexical variation, are word-level properties of the training phrases and are not directly readable from the log event counts, so they remain independent predictions. I found no load-bearing self-citation chain: references to Borchers et al. for timestamp alignment and to prior work on log and dialogue modeling are methodological, not circular proof. The outcome validation against expert ratings and end-user satisfaction is external. Overall circularity is partial, confined to the productivity outcome and the associated fusion claim, so the score is 6.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the validity of the outcome proxies and on the train/test split properly isolating pairs. The free parameters are mostly standard ML hyperparameters plus one data preprocessing cutoff.

free parameters (3)
  • Neural network hyperparameters (hidden layer sizes, dropout rate) = not reported per outcome
    Selected via grid search over the specified parameter space with 5-fold CV; the best configuration per outcome is not reported.
  • Segment length threshold = 12.5 minutes (90th percentile)
    Ad-hoc cutoff to remove excessively long intent-working segments; affects the dataset composition.
  • Early stopping patience = 2 epochs
    Chosen to prevent overfitting.
assumptions (4)
  • domain assumption The three outcome measures (training phrase count, lexical density, lexical variation) are valid proxies for project quality
    The paper validates via learning curves and correlations with final project scores, but the constructs remain assumptions.
  • domain assumption Student-level five-fold cross-validation prevents data leakage between training and test folds
    The paper does not state whether both members of a collaborating pair are kept in the same fold; if not, pair-level dialogue/log context could leak.
  • domain assumption Transcript-log timestamp alignment to 1 second is accurate
    Section 3.3 states alignment accuracy of 1 second following prior work [18], but the alignment errors are not quantified.
  • ad hoc to paper Removing segments longer than 12.5 minutes does not introduce bias
    The authors argue the outcome measures are uncorrelated with dialogue length, but this is a post-hoc exclusion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Combining Log Data and Collaborative Dialogue Features to Predict Project Quality in Middle School AI Education." pith.science (2026). https://pith.science/paper/TNNT267A

@misc{pith2026250611326,
  author       = {Pith},
  title        = {Pith review of: Combining Log Data and Collaborative Dialogue Features to Predict Project Quality in Middle School AI Education},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TNNT267A}},
  note         = {Machine review of arXiv:2506.11326}
}
read the original abstract

Project-based learning plays a crucial role in computing education. However, its open-ended nature makes tracking project development and assessing success challenging. We investigate how dialogue and system interaction logs predict project quality during collaborative, project-based AI learning of 94 middle school students working in pairs. We used linguistic features from dialogue transcripts and behavioral features from system logs to predict three project quality outcomes: productivity (number of training phrases), content richness (word density), and lexical variation (word diversity) of chatbot training phrases. We compared the predictive accuracy of each modality and a fusion of the modalities. Results indicate log data better predicts productivity, while dialogue data is more effective for content richness. Both modalities modestly predict lexical variation. Multimodal fusion improved predictions for productivity and lexical variation of training phrases but not content richness. These findings suggest that the value of multimodal fusion depends on the specific learning outcome. The study contributes to multimodal learning analytics by demonstrating the nuanced interplay between behavioral and linguistic data in assessing student learning progress in open-ended AI learning environments.

Figures

Figures reproduced from arXiv: 2506.11326 by the authors.

Figure 1
Figure 1. Chatbot development environment AMBY a partner at the same computer on iterating on the chatbot design, following the pair programming paradigm [42], prompting spontaneous collaboration, dialogue, and joint decision-making. The system collects 23 types of timestamped user interaction logs (e.g., adding or deleting training phrases, submitting AI training requests, chat messages). Across 121 recorded collaborative se… view at source ↗
Figure 2
Figure 2. Overview of our multimodal predictive model, an example dataset and project outcome substantive content over filler words [43, 44]. Lexical Variation measures language diversity through the ratio of unique content words to total content words [43], important because varied phrases help chatbots recognize different user expressions. Higher variation indicates more robust training input. These project quality metrics … view at source ↗
Figure 3
Figure 3. Learning curves for training phrase count, lexical density, and lexical variation, showing student improvements over time through smoothed running averages. Trends were less consistent for students with high submission counts (>30 intents vs. average 17), likely representing outliers [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 56 canonical work pages

  1. [1]

    Pucher, M

    R. Pucher, M. Lehner, Project based learning in computer science–a review of more than 500 projects, Procedia-Social and Behavioral Sciences 29 (2011) 1561–1566

    work page 2011

  2. [2]

    Kokotsaki, V

    D. Kokotsaki, V. Menzies, A. Wiggins, Project-based learning: A review of the literature, Improving schools 19 (2016) 267–277

    work page 2016

  3. [3]

    J. H. L. Koh, S. C. Herring, K. F. Hew, Project-based learning and student knowledge construction during asynchronous online discussion, The Internet and Higher Education 13 (2010) 284–291

    work page 2010

  4. [4]

    P. Guo, N. Saab, L. S. Post, W. Admiraal, A review of project-based learning in higher education: Student outcomes and measures, International journal of educational research 102 (2020) 101586

    work page 2020

  5. [5]

    S. W. Widyaningsih, I. Yusuf, Implementation of project-based learning (pjbl) assisted by e-learning through lesson study activities to improve the quality of learning in physics learning planning courses., International Journal of Higher Education 9 (2020) 60–68

    work page 2020

  6. [6]

    M. H. Wilkerson-Jerde, Construction, categorization, and consensus: Student generated compu- tational artifacts as a context for disciplinary reflection, Educational Technology Research and Development 62 (2014) 99–121

    work page 2014

  7. [7]

    Oviatt, A

    S. Oviatt, A. Cohen, Written and multimodal representations as predictors of expertise and problem-solving success in mathematics, in: Proceedings of the 15th ACM on International conference on multimodal interaction, 2013, pp. 599–606

    work page 2013

  8. [8]

    J. Yoo, J. Kim, Can online discussion participation predict group project performance? investigating the roles of linguistic features and participation patterns, International Journal of Artificial Intelligence in Education 24 (2014) 8–32

    work page 2014

Show all 60 references
  1. [9]

    X. Tian, A. Mannekote, C. E. Solomon, Y. Song, C. F. Wise, T. Mcklin, J. Barrett, K. E. Boyer, M. Israel, Examining llm prompting strategies for automatic evaluation of learner-created computational artifacts, in: Proceedings of the International Conference on Educational Data...

    2024

  2. [10]

    M. A. Samadi, N. Nixon, Cultural diversity in team conversations: A deep dive into its effects on cohesion and team performance, in: Proceedings of the International Conference on Educational Data Mining (EDM), 2024, pp. 821–827

    2024

  3. [11]

    A. E. Stewart, Z. Keirn, S. K. D’Mello, Multimodal modeling of collaborative problem-solving facets in triads, User Modeling and User-Adapted Interaction 31 (2021) 713–751

    2021

  4. [12]

    Acosta, S

    H. Acosta, S. Lee, B. Mott, H. Bae, K. Glazewski, C. Hmelo-Silver, J. Lester, Multimodal learning analytics for predicting student collaboration satisfaction in collaborative game-based learning (2024)

    2024

  5. [13]

    X. Tian, A. E. Griffith, Z. Price, K. E. Boyer, K. Tang, Investigating linguistic alignment in collabo- rative dialogue: A study of syntactic and lexical patterns in middle school students, Language and Speech (2024) 00238309241234565

    2024

  6. [14]

    Grafsgaard, J

    J. Grafsgaard, J. B. Wiggins, K. E. Boyer, E. N. Wiebe, J. Lester, Automatically recognizing facial expression: Predicting engagement and frustration, in: Proceedings of International Conference on Educational Data Mining (EDM), 2013

    2013

  7. [15]

    Morshed Fahid, X

    F. Morshed Fahid, X. Tian, A. Emerson, J. B. Wiggins, D. Bounajim, A. Smith, E. Wiebe, B. Mott, K. Elizabeth Boyer, J. Lester, Progression trajectory-based student modeling for novice block-based programming, in: Proceedings of the 29th ACM Conference on User Modeling, Adaptat...

    2021

  8. [16]

    Marwan, J

    S. Marwan, J. Jay Williams, T. Price, An evaluation of the impact of automated programming hints on performance and learning, in: Proceedings of the 2019 ACM Conference on International Computing Education Research, 2019, pp. 61–70

    2019

  9. [17]

    Emerson, F

    A. Emerson, F. J. Rodríguez, B. Mott, A. Smith, W. Min, K. E. Boyer, C. Smith, E. Wiebe, J. Lester, Pre- dicting early and often: Predictive student modeling for block-based programming environments., Proceedings of International Conference on Educational Data Mining (EDM) (2019)

    2019

  10. [18]

    Borchers, J

    C. Borchers, J. Zhang, R. S. Baker, V. Aleven, Using think-aloud data to understand relations between self-regulation cycle characteristics and student performance in intelligent tutoring systems, in: Proceedings of the 14th Learning Analytics and Knowledge Conference, 2024, p...

    2024

  11. [19]

    Borchers, K

    C. Borchers, K. Yang, J. Lin, N. Rummel, K. R. Koedinger, V. Aleven, Combining dialog acts and skill modeling: What chat interactions enhance learning rates during ai-supported peer tutoring?, in: Proceedings of the International Conference on Educational Data Mining (EDM), 2024

    2024

  12. [20]

    Zhang, C

    J. Zhang, C. Borchers, V. Aleven, R. S. Baker, Using large language models to detect self-regulated learning in think-aloud protocols, in: Proceedings of the International Conference on Educational Data Mining (EDM), 2024

    2024

  13. [21]

    Spikol, E

    D. Spikol, E. Ruffaldi, M. Cukurova, Using multimodal learning analytics to identify aspects of collaboration in project-based learning, in: CSCL’17: The 12th International Conference on Computer Supported Collaborative Learning, volume 1, International Society of the Learning...

    2017

  14. [22]

    L. Yan, V. Echeverria, Y. Jin, G. Fernandez-Nieto, L. Zhao, X. Li, R. Alfredo, Z. Swiecki, D. Gašević, R. Martinez-Maldonado, Evidence-based multimodal learning analytics for feedback and reflection in collaborative learning, British Journal of Educational Technology 55 (2024)...

    2024

  15. [23]

    J. K. Olsen, K. Sharma, N. Rummel, V. Aleven, Temporal analysis of multimodal data to predict collaborative learning outcomes, British Journal of Educational Technology 51 (2020) 1527–1547

    2020

  16. [24]

    Sharma, M

    K. Sharma, M. Giannakos, Multimodal data capabilities for learning: What can multimodal data tell us about learning?, British Journal of Educational Technology 51 (2020) 1450–1484

    2020

  17. [25]

    Karumbaiah, C

    S. Karumbaiah, C. Borchers, T. Shou, A.-C. Falhs, P. Liu, T. Nagashima, N. Rummel, V. Aleven, A spatiotemporal analysis of teacher practices in supporting student learning and engagement in an ai-enabled classroom, in: International Conference on Artificial Intelligence in Edu...

    2023

  18. [26]

    Chango, J

    W. Chango, J. A. Lara, R. Cerezo, C. Romero, A review on data fusion in multimodal learning ana- lytics and educational data mining, Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 12 (2022) e1458

    2022

  19. [27]

    Scarlatos, A

    A. Scarlatos, A. Lan, Exploring knowledge tracing in tutor-student dialogues, arXiv preprint arXiv:2409.16490 (2024)

    2024 arXiv

  20. [28]

    Walker, N

    E. Walker, N. Rummel, K. R. Koedinger, Adaptive intelligent support to improve peer tutoring in algebra, International Journal of Artificial Intelligence in Education 24 (2014) 33–61

    2014

  21. [29]

    Costley, How system functionality improves the effectiveness of collaborative learning, Interac- tive Learning Environments 30 (2022) 971–983

    J. Costley, How system functionality improves the effectiveness of collaborative learning, Interac- tive Learning Environments 30 (2022) 971–983

    2022

  22. [30]

    Roschelle, S

    J. Roschelle, S. D. Teasley, The construction of shared knowledge in collaborative problem solving, in: Computer supported collaborative learning, Springer, 1995, pp. 69–97

    1995

  23. [31]

    Echeverria, L

    V. Echeverria, L. Yan, L. Zhao, S. Abel, R. Alfredo, S. Dix, H. Jaggard, R. Wotherspoon, A. Osborne, S. Buckingham Shum, et al., Teamslides: a multimodal teamwork analytics dashboard for teacher- guided reflection in a physical learning space, in: Proceedings of the 14th Learn...

    2024

  24. [32]

    Q. Zhou, W. Suraworachet, M. Cukurova, Detecting non-verbal speech and gaze behaviours with multimodal data and computer vision to interpret effective collaborative learning interactions, Education and Information Technologies 29 (2024) 1071–1098

    2024

  25. [33]

    Abdelshiheed, J

    M. Abdelshiheed, J. K. Jacobs, S. K. D’Mello, Aligning tutor discourse supporting rigorous think- ing with tutee content mastery for predicting math achievement, in: Proceedings of the 25th International Conference on Artificial Intelligence in Education (AIED’24), Recife, Bra...

    2024

  26. [34]

    A. M. Latham, K. A. Crockett, D. A. McLean, B. Edmonds, K. O’shea, Oscar: An intelligent conversational agent tutor to estimate learning styles, in: International conference on fuzzy systems, IEEE, 2010, pp. 1–8

    2010

  27. [35]

    Matsuda, E

    N. Matsuda, E. Yarzebinski, V. Keiser, R. Raizada, G. J. Stylianides, W. W. Cohen, K. R. Koedinger, Learning by teaching simstudent–an initial classroom baseline study comparing with cognitive tutor, in: International Conference on Artificial Intelligence in Education (AIED 20...

    2011

  28. [36]

    X. Tian, N. Lubold, L. Friedman, E. Walker, Understanding rapport over multiple sessions with a social, teachable robot, in: International Conference on Artificial Intelligence in Education (AIED 2020), Springer, 2020, pp. 318–323

    2020

  29. [37]

    Weerasinghe, A

    A. Weerasinghe, A. Mitrovic, M. Van Zijl, B. Martin, Evaluating the effectiveness of adaptive tutorial dialogues in database design, in: Proceedings of the 18th International Conference on Computers in Education, 2010, pp. 33–40

    2010

  30. [38]

    D’Mello, A

    S. D’Mello, A. Olney, C. Williams, P. Hays, Gaze tutor: A gaze-reactive intelligent tutoring system, International Journal of Human-Computer Studies 70 (2012) 377–398

    2012

  31. [39]

    Y. Song, G. A. Katuka, J. Barrett, X. Tian, A. Kumar, T. McKlin, M. Celepkolu, K. E. Boyer, M. Is- rael, Ai made by youth: A conversational ai curriculum for middle school summer camps, in: Proceedings of the Thirty-Seventh AAAI Conference on Artificial Intelligence and Thirty...

    2023

  32. [40]

    Touretzky, C

    D. Touretzky, C. Gardner-McCune, F. Martin, D. Seehorn, Envisioning ai for k-12: What should every child know about ai?, in: Proceedings of the AAAI conference on artificial intelligence, volume 33, 2019, pp. 9795–9799

    2019

  33. [41]

    X. Tian, A. Kumar, C. E. Solomon, K. D. Calder, G. A. Katuka, Y. Song, M. Celepkolu, L. Pezzullo, J. Barrett, K. E. Boyer, et al., Amby: A development environment for youth to create conversational agents, International Journal of Child-Computer Interaction 38 (2023) 100618

    2023

  34. [42]

    Hanks, S

    B. Hanks, S. Fitzgerald, R. McCauley, L. Murphy, C. Zander, Pair programming in education: A literature review, Computer Science Education 21 (2011) 135–173

    2011

  35. [43]

    Kim, Predicting l2 writing proficiency using linguistic complexity measures: A corpus-based study., English teaching 69 (2014)

    J.-y. Kim, Predicting l2 writing proficiency using linguistic complexity measures: A corpus-based study., English teaching 69 (2014)

    2014

  36. [44]

    Lu, The relationship of lexical richness to the quality of esl learners’ oral narratives, The Modern Language Journal 96 (2012) 190–208

    X. Lu, The relationship of lexical richness to the quality of esl learners’ oral narratives, The Modern Language Journal 96 (2012) 190–208

    2012

  37. [45]

    W. Park, H. Kwon, Implementing artificial intelligence education for middle school technology education in republic of korea, International Journal of Technology and Design Education 34 (2023) 109–135. doi:10.1007/s10798-023-09812-2

    2023 doi

  38. [46]

    Lakkaraju, T

    K. Lakkaraju, T. Hassan, V. Khandelwal, P. Singh, C. Bradley, R. Shah, F. Agostinelli, B. Srivastava, D. Wu, Allure: a multi-modal guided environment for helping children learn to solve a rubik’s cube with automatic solving and interactive explanations, Proceedings of the AAAI...

    2022 doi

  39. [47]

    Yu, Lexical diversity in writing and speaking task performances, Applied linguistics 31 (2010) 236–259

    G. Yu, Lexical diversity in writing and speaking task performances, Applied linguistics 31 (2010) 236–259

    2010

  40. [48]

    B. Kondal, Effects of lexical density and lexical variety in language performance and proficiency, International Journal of IT, Engineering and Applied Sciences Research (IJIEASR) 4 (2015) 25–29

    2015

  41. [49]

    Rivers, E

    K. Rivers, E. Harpstead, K. R. Koedinger, Learning curve analysis for programming: Which concepts do students struggle with?, in: Proceedings of 2016 ACM Conference on International Computing Education Research, volume 16, ACM, 2016, pp. 143–151

    2016

  42. [50]

    M. Chi, K. Koedinger, G. Gordon, P. Jordan, Instructional factors analysis: A cognitive model for multiple instructional interventions, in: Proceedings of the 4th International Conference on Educational Data Mining (EDM), 2011

    2011

  43. [51]

    J. Lin, S. Singh, L. Sha, W. Tan, D. Lang, D. Gašević, G. Chen, Is it a good move? mining effective tutoring strategies from human–human tutorial dialogues, Future Generation Computer Systems 127 (2022) 194–207

    2022

  44. [52]

    Reimers, Sentence-bert: Sentence embeddings using siamese bert-networks, arXiv preprint arXiv:1908.10084 (2019)

    N. Reimers, Sentence-bert: Sentence embeddings using siamese bert-networks, arXiv preprint arXiv:1908.10084 (2019)

    2019 arXiv

  45. [53]

    L. Wang, N. Yang, X. Huang, L. Yang, R. Majumder, F. Wei, Improving text embeddings with large language models, arXiv preprint arXiv:2401.00368 (2023)

    2023 arXiv

  46. [54]

    Grover, R

    S. Grover, R. Pea, Computational thinking in k–12: A review of the state of the field, Educational researcher 42 (2013) 38–43

    2013

  47. [55]

    Condor, Z

    A. Condor, Z. Pardos, Explainable automatic grading with neural additive models, in: International Conference on Artificial Intelligence in Education, Springer, 2024, pp. 18–31

    2024

  48. [56]

    Hessel, L

    J. Hessel, L. Lee, Does my multimodal model learn cross-modal interactions? it’s harder to tell than you might think!, Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) (2020). doi:10.18653/v1/2020.emnlp-main.62

    2020 doi

  49. [57]

    K. B. Yang, C. Borchers, A.-C. Falhs, V. Echeverria, S. Karumbaiah, N. Rummel, V. Aleven, Leverag- ing multimodal classroom data for teacher reflection: Teachers’ preferences, practices, and privacy considerations, in: European Conference on Technology Enhanced Learning, Sprin...

    2024

  50. [58]

    Ma, Multimodal Modeling of Collaborative Learning with Adaptive Data Fusion, Ph.D

    Y. Ma, Multimodal Modeling of Collaborative Learning with Adaptive Data Fusion, Ph.D. thesis, 2023

    2023

  51. [59]

    C. Cohn, E. Davalos, C. Vatral, J. H. Fonteles, H. D. Wang, M. Ma, G. Biswas, Multimodal methods for analyzing learning and training environments: A systematic literature review, arXiv preprint arXiv:2408.14491 (2024)

    2024

  52. [60]

    Wixon, R

    M. Wixon, R. S. d. Baker, J. D. Gobert, J. Ocumpaugh, M. Bachmann, Wtf? detecting students who are conducting inquiry without thinking fastidiously, in: User Modeling, Adaptation, and Personalization: 20th International Conference, UMAP 2012, Montreal, Canada, July 16-20, 2012...

    2012

Pith tools

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