Pith. sign in

REVIEW 5 major objections 7 minor 41 references

Beyond Instructed Tasks: Recognizing In-the-Wild Reading Behaviors in the Classroom Using Eye Tracking

T0 review · 5 major / 7 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read Real classroom reading differs from instructed reading, and an image-based neural network can spot three natural reading behaviors from ten-fixation gaze scanpaths with a macro F1 of 0.80.

desk verdict Genuine proof-of-concept for in-the-wild reading behavior recognition, but the F1=0.80 rests on moderately reliable human labels and needs an honest limitations section. read the letter →

arxiv 2501.18468 v1 pith:XLPRNAKN submitted 2025-01-30 cs.HC cs.AI

classification cs.HCcs.AI
keywords readingbehaviorrecognitioneyetrackingin-the-wildclassroomstudyscanpathimagestaxonomyconvolutionalneuralnetworkreal-timeclassification
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 asks whether reading behaviors can be recognized from eye gaze in a real classroom, without instructing students how to read. It reports that natural reading produces gaze patterns that differ measurably from instructed reading, and it organizes those patterns into six behaviors defined by velocity, gaze density, and sequentiality. The central empirical claim is that a lightweight 2D convolutional network, fed with images of ten-fixation gaze scanpaths, can classify three of those behaviors—sequential, non-sequential, and skimming—with a macro F1 of 0.80 under leave-one-participant-out cross-validation. If that claim holds, real-time feedback about how students read is feasible on ordinary classroom hardware, although the study is explicitly a proof of concept rather than a generalizable deployment.

What carries the argument

The central mechanism is the scanpath image: a rendered picture of a ten-fixation gaze window over the text, colored blue-to-red by time, which lets a 2D convolutional network see the shape and direction of reading without hand-engineered features. Supporting this, the paper defines a velocity–density–sequentiality framework: words per minute, inverse fixation dispersion, and forward-versus-backward saccade ratio, which human annotators used to define the behavior taxonomy and which the statistical analyses show separates the classes. The data pipeline converts raw gaze points into page coordinates in real time, so scrolling and zooming do not corrupt the scanpath image. The classifier stacks a pretrained residual-network backbone with three fully connected layers and runs at about 3 ms on a GPU and 14 ms on a CPU, far below the roughly six-second input window.

What would settle it

Take the same gaze videos and have two fresh, independent annotation teams label them under the same protocol: if their agreement is again only moderate (kappa around 0.57–0.65), and a classifier trained on one team's labels does no better than chance on the other team's labels, then the F1 of 0.80 reflects the specific annotators rather than stable reading behaviors.

Watch

Extended reading notes

Core claim

The paper's central discovery is that in-the-wild reading behavior in a classroom can be defined and automatically detected from gaze data alone, and that it is not the same as behavior performed on instruction. Human reviewers watching replay videos of gaze scanpaths from 27 sixth-grade students identified six archetypes—static, deep, sequential, non-sequential, skimming, and previewing/mapping—and the paper shows these are distinguishable along three continua: words per minute (velocity), inverse fixation dispersion (density), and the forward-versus-backward saccade ratio (sequentiality). Pairwise multivariate tests separate most behavior pairs, with only skimming versus previewing/mapping not reaching significance. Instructed and in-the-wild gaze distributions differ significantly for sequential and deep reading but not for skimming. A lightweight 2D convolutional network using rendered ten-fixation scanpath images as input reaches a macro F1 of 0.80 (precision 0.82, recall 0.81) for sequential, non-sequential, and skimming under leave-one-participant-out cross-validation, with inference fast enough for real-time use.

Load-bearing premise

The result rests on the assumption that the behaviors named by the human annotators are stable, real categories rather than subjective interpretations of noisy gaze videos, because both the statistical separations and the classifier's F1 score are measured against those labels.

Editorial extensions

If this is right

  • Instructed reading tasks are not a safe stand-in for real classroom reading, so behavior models meant for classrooms should be trained on naturalistic gaze data.
  • A ten-fixation window can capture short behaviors such as skims that longer sliding-window methods (over 30 seconds) would blend into multiple labels.
  • The pipeline runs faster than real time (about 3 ms inference on a GPU and 14 ms on a CPU), so per-section behavior feedback during a lesson is technically plausible.
  • Mapping detected behaviors to passage locations could let teachers see which sections a student skimmed, read sequentially, or revisited, and connect those patterns to comprehension questions.

Reading between the lines

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

  • If velocity, density, and sequentiality are true axes of reading behavior, the same ten-fixation scanpath representation should transfer to other texts, age groups, and languages; a direct transfer test would reveal whether the taxonomy is universal or specific to this passage and grade.
  • Because the classifier is trained on human labels with only moderate inter-rater agreement, its practical ceiling is the reliability of the annotation protocol; checking predicted labels against an independent outcome, such as correctness on comprehension questions tied to specific text sections, would separate behavior recognition from annotator subjectivity.
  • The statistical overlap between skimming and previewing/mapping suggests the taxonomy may be treating two variants of one fast behavior as two categories; more instances of previewing/mapping could either sharpen the boundary or merge the labels.
  • The finding that skimming looks similar under instruction and in the wild hints that skimming is a more stable reading mode, while sequential and deep reading are more context-sensitive; this could make skimming a better target for general reading instruction.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 7 minor

Summary. The paper presents a classroom eye-tracking study with 27 sixth-grade students, combining instructed-reading and in-the-wild reading conditions. The authors develop a six-behavior taxonomy ('static', 'deep', 'sequential', 'non-sequential', 'skimming', 'previewing/mapping') through iterative human annotation, propose a three-dimension framework (velocity, density, sequentiality), report statistical differences between instructed and in-the-wild behaviors and between behavior pairs, and train a 2D CNN on 10-fixation scanplot images to classify three of the six behaviors, reporting a macro F1 of 0.80 under leave-one-participant-out cross-validation. The central claims are that in-the-wild reading behaviors differ measurably from instructed ones and that a lightweight CNN can recognize such behaviors in real time.

Significance. If the reported results hold, the paper would make a useful contribution: the ecological classroom dataset, the explicit behavior taxonomy, and the real-time classification pipeline are all potentially valuable to the reading-behavior and educational-technology communities. The use of leave-one-participant-out cross-validation and the reporting of inference times are strengths, and the planned public release of code and data would support reproducibility. However, the central empirical claim currently rests on human annotations whose reliability is moderate (Cohen's Kappa 0.57–0.65), and several validation steps are circular or incompletely specified. The contribution is therefore best viewed as a proof-of-concept whose scope and limitations must be stated more precisely before the claims can be accepted as stated.

major comments (5)
  1. [Sections 4.1, 5.3, 5.4] The ground truth for both the statistical comparisons and the CNN is the set of human labels produced in the iterative annotation process described in Section 4.1. Section 4.1.2 reports that the taxonomy was created and renamed during the annotation rounds, and that final Cohen's Kappa values are 0.57 and 0.65 (Table 2), which are moderate. Because the same labels are used to train the CNN and to run the Hotelling T-squared and t-tests, the reported F1 and p-values measure the separability of the annotators' own judgments rather than the separability of objectively defined reading states. The paper should either validate the labels with an independent, codebook-based annotation by new coders or explicitly limit all conclusions to 'human-defined behavior classes' and add annotation reliability to Section 8's limitation list.
  2. [Sections 4.1.1, 4.2.2, Fig. 8] The 2D CNN is trained on scanplot images rendered with the same blue-to-red temporal gradient that the human reviewers used when annotating the gaze scanpath videos (see the caption of Fig. 8 and Section 4.1.1). The macro F1 of 0.80 may therefore reflect the network learning the reviewers' visual heuristics rather than recognizing stable reading states. To support the claim of a 'robust approach for understanding in-the-wild reading', the authors should demonstrate that the model generalizes to labels obtained from a fixed codebook or that its predictions correlate with an independent behavioral outcome.
  3. [Abstract and Section 4.2] The abstract states that the 2D CNN achieved an F1 score of 0.8 'for behavior recognition' and that the framework differentiates reading behaviors, but Section 4.2 explicitly restricts the classifier to three of the six defined behaviors ('sequential', 'non-sequential', and 'skimming') because of class imbalance and limited instances of the other behaviors. All abstract and discussion claims should be qualified to say that the F1 score applies to three behavior classes under the study's narrow conditions, not to the full six-behavior taxonomy.
  4. [Table 14a, Section 5.4] The baseline results in Table 14a are internally inconsistent under standard definitions. For a three-class problem with a dominant 'sequential' class, an always-majority classifier has macro F1 at most 1/3 in the usual macro-average convention (for a sequential proportion around 0.87, macro F1 is approximately 0.31), yet the table reports 0.56 for both 'Majority Class' and 'SVC'. Similarly, a random classifier with class-prior probabilities would have macro F1 around 0.3 and accuracy around 0.76, not the reported 0.38 and 0.79. Please specify exactly how 'Random' and 'Majority Class' baselines were constructed and report class-wise precision/recall or confusion matrices for all baselines.
  5. [Section 4.2.2, Section 5.4] The exhaustive grid search over learning rate, weight decay, model design, batch size, and scanpath plot generation is described without stating how hyperparameters were selected relative to the leave-one-participant-out folds. If the grid search used the same test folds that produce the reported F1, the performance estimate is optimistically biased. Please describe the inner validation procedure and report the variance or confidence intervals of the macro F1 across the 27 folds.
minor comments (7)
  1. [Section 2.3] There is a typo in 'readi recognition' that should read 'reading recognition'.
  2. [Section 5.1] The phrase 'p < 0.348' for the skimming comparison should be 'p = 0.348' or similar, since an inequality with a non-significant p-value is not standard.
  3. [Section 5.3] The cross-reference 'as shown in Fig. 5' should refer to Table 5, which contains the pairwise Hotelling T-squared p-values.
  4. [Section 5.4] The caption 'Table 14a' is used for a table and 'Fig. 14b' for a confusion matrix; the numbering of tables and figures should be separated and consistent with the journal style.
  5. [Section 4.2.2] There is a grammatical error: 'Gaussian noise was add' should be 'Gaussian noise was added'.
  6. [Section 6.3] The text says a fixed 10-fixation window corresponds to 6 seconds, but with typical fixation durations of 200–300 ms this would be 2–3 seconds; please clarify how the 6-second estimate is derived.
  7. [Section 8] The limitations section does not mention the moderate inter-rater reliability or the lack of independent annotation validation; adding this would give readers an accurate picture of the strength of the ground-truth labels.

Circularity Check

1 steps flagged · score 4.0 of 10

Taxonomy validation is partially self-definitional; the CNN F1 remains an empirical, annotation-relative result rather than a circular derivation.

  1. self definitional [Section 5.2 (Theoretical Framework and Taxonomy) and Section 5.3 (Velocity, Density, and Sequentiality Continua)]
    "Throughout our human-driven annotation process, the language we used to resolve disagreements and describe reading behaviors was instrumental in developing the framework presented in Fig. 12. Reviewers relied on key variables, such as velocity, density, and sequentiality of the gaze scanpaths, when assigning behavior labels and reaching consensus in cases of disagreement. ... The similarity of the jointplot to the theoretical framework and the results of the significance testing further support our behavior regions shown in Fig. 12a."

    The framework is induced from the annotators' use of velocity, density, and sequentiality to assign labels, and the validation then uses the same three continua on the same labeled segments: pairwise Hotelling T-squared tests on WPM and inverse fixation dispersion, and an independent t-test on FBSR. The statistical separation is therefore largely built into the labeling protocol rather than providing independent confirmation of the taxonomy. It demonstrates internal consistency of the annotation process, not an independent derivation of the behavior categories.

full rationale

The paper's most prominent empirical result, the 2D CNN macro F1 of 0.80 under leave-one-participant-out cross-validation, is a standard supervised classification evaluation against human-annotated labels. It has real empirical content because the model sees held-out participants and must generalize from scanplot images to labels; it does not reduce to a fitted parameter, a self-citation chain, or an identical restatement of its inputs. However, the validation of the theoretical framework is partially circular: the taxonomy was created by reviewers who explicitly used velocity, density, and sequentiality to label gaze scanpaths, and Section 5.3 then treats statistical differences on those same variables as support for the framework. Moderate inter-rater reliability (Cohen's Kappa 0.57-0.65) and the absence of independent reproducibility checks for the final labels are important validity limitations, but they are not circularity per se. No load-bearing self-citations or imported uniqueness arguments are present. Overall, the central classifier result survives as an empirical measure of agreement with human judgments, but the framework-validation step is substantially self-definitional, giving a moderate overall circularity score.

Assumptions & free parameters 5 free parameters · 4 assumptions · 3 invented entities

The paper pulls most of its analytical structure from the prior literature (Campbell and Maglio, Busjahn, Srivastava) but introduces several hand-set thresholds and a de novo taxonomy. The CNN adds trained weights as effective free parameters, though their exact values are not reported. The main epistemic burden is that the behavior labels are defined by the authors from the same data used to validate the framework.

free parameters (5)
  • fixation_window_length = 10 fixations
    Chosen to balance granularity and context; affects all CNN predictions.
  • static behavior threshold = approximately 5 seconds
    Hand-set by the annotators to label 'static' behavior; a free choice that shapes the taxonomy.
  • deep behavior threshold = 3 or more repeated line traversals
    Hand-set criterion for labeling 'deep' reading in the annotation protocol.
  • CNN hyperparameters = not reported
    Learning rate, weight decay, batch size, and scanplot generation were selected by exhaustive grid search; exact values are not given in the paper.
  • time-window sizes for ML baselines = t = 2 to 15 seconds
    Range chosen based on the per-behavior duration distribution; the SVC baseline uses t=15.
assumptions (4)
  • domain assumption Segments from the same participant are treated as independent samples in the statistical tests.
    The Mann-Whitney U, t-test, and Hotelling T-squared tests do not account for clustering by participant. This enters in Section 5, where behavior segments are compared without a mixed-effects or repeated-measures correction.
  • domain assumption The eye-to-page coordinate transformation (Equation 1) accurately maps gaze to the PDF content despite scrolling and zooming.
    The transformation relies on the DOM bounding rectangle of the page element. No validation of this mapping against ground-truth page positions is reported.
  • domain assumption Scanplot images with a blue-to-red temporal gradient encode enough temporal ordering for a CNN to learn reading behaviors.
    The 2D CNN input is a rendered image of the scanpath; the paper does not demonstrate that this representation is a sufficient statistic for velocity, density, and sequentiality.
  • ad hoc to paper The behavior taxonomy defined by the authors is a valid ontological classification of reading behaviors.
    Labels such as non-sequential, previewing/mapping, and static were invented by the authors from observations of the same dataset, without external validation against, for example, comprehension outcomes.
invented entities (3)
  • Non-sequential reading behavior label
    purpose: Captures a scanpath pattern that does not follow the conventional line-by-line order but is not a full skim.
    This label was defined by the authors from their own gaze data and then used to train and evaluate the classifier, so it lacks independent validation.
  • Previewing/mapping behavior label
    purpose: Describes brief, large-saccade gaze movements that appear to map out document layout.
    Observed only in the paper's dataset; no external behavioral measure is linked to it.
  • Static behavior label
    purpose: Represents gaze remaining nearly stationary for more than about 5 seconds.
    The threshold is arbitrary and the label is only based on the authors' visual inspection of scanpaths.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Instructed Tasks: Recognizing In-the-Wild Reading Behaviors in the Classroom Using Eye Tracking." pith.science (2026). https://pith.science/paper/XLPRNAKN

@misc{pith2026250118468,
  author       = {Pith},
  title        = {Pith review of: Beyond Instructed Tasks: Recognizing In-the-Wild Reading Behaviors in the Classroom Using Eye Tracking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XLPRNAKN}},
  note         = {Machine review of arXiv:2501.18468}
}
read the original abstract

Understanding reader behaviors such as skimming, deep reading, and scanning is essential for improving educational instruction. While prior eye-tracking studies have trained models to recognize reading behaviors, they often rely on instructed reading tasks, which can alter natural behaviors and limit the applicability of these findings to in-the-wild settings. Additionally, there is a lack of clear definitions for reading behavior archetypes in the literature. We conducted a classroom study to address these issues by collecting instructed and in-the-wild reading data. We developed a mixed-method framework, including a human-driven theoretical model, statistical analyses, and an AI classifier, to differentiate reading behaviors based on their velocity, density, and sequentiality. Our lightweight 2D CNN achieved an F1 score of 0.8 for behavior recognition, providing a robust approach for understanding in-the-wild reading. This work advances our ability to provide detailed behavioral insights to educators, supporting more targeted and effective assessment and instruction.

Figures

Figures reproduced from arXiv: 2501.18468 by the authors.

Figure 1
Figure 1. In-the-Wild Reading Behaviors in Naturalistic Hierarchical Passages: Displaying the human-labeled reading behavior annotations onto an authentic reading passage. Understanding reader behaviors such as skimming, deep reading, and scanning is essential for improving educational instruction. While prior eye-tracking studies have trained models to recognize reading behaviors, they often rely on instructed reading tasks,… view at source ↗
Figure 2
Figure 2. Study Structure: The study is organized into three main sections: setup, instructed behavior activity, and in-the-wild reading. During the setup phase, students completed an eye-tracking calibration and received instructions. In the instructed behavior activity, students were directed to read in a specified manner. For the in-the-wild reading section, students were asked to read a passage as they would naturally in … view at source ↗
Figure 3
Figure 3. Data Collection Protocol: Deployed using consumer-level HP laptops to present the custom web application and a Tobii Pro Spark to collect eye-tracking data. The web application provides a digital PDF reader with a side panel for questions. 3.3 Preprocessing With the recorded gaze data, our goal was to analyze the gaze scanpath on the PDF content to understand reading behaviors and how students approached text compre… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Page AOI Encoding: Illustration of the relationship between screen coordinate space (SCS) and an example page coordinate space (PCS) of the PDF viewer within the custom web application [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Example gaze scanpath: Utilizing a blue-to-red gradient to illustrate the temporal evaluation of the scanpath. The visualization is overlaid on top of the text to help contextualize the gaze. The beginning of the scanpath is painted blue (0, 0, 255), throughout time th…
Figure 6
Figure 6. Figure 6: Example Human-Labeled In-the-Wild Behaviors: Gaze scanpaths from participants 1 (top), 15 (center), and 47 (bottom). illustrating how different reading behaviors are leveraged to explore and comprehend the text. 10 [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: In-the-wild Behavior Corpus Characteristics: Duration, count, and fixation count for accumulative and per-behavior. From these plots, we can observe the significant class imbalance favoring “sequential” reading. Lastly, we visualize the duration and fixation length dis…
Figure 8
Figure 8. Figure 8: Model Input Examples of 2D Gaze Scanpaths: “sequential” reading on the left, “non-sequential” reading on the center, and "skim” reading on the right. These scanpath input images are generated from a fixation window that is rendered on a standard letter-size image. Simi…
Figure 9
Figure 9. Figure 9: Examples of Gaze Scanpaths for Instructed Behavior: “deep” reading on the left, “regular” reading on the center, and “skim” reading on the right. The box plots in [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: In-the-Wild Behavior Archetypes: During our human-driven annotation process, recorded examples of the behaviors as a reference to have a visual representation to compare against new behaviors. Example scanpath plots of each in-the-wild behavior, with “static” having a…
Figure 11
Figure 11. Figure 11: Gaze Measure Differences Between Instructed and In-the-Wild Behaviors: The velocity (left) and density (right distributions illustrate how instructed behaviors do not match in-the-wild behaviors. 5.2 Theoretical Framework and Taxonomy Throughout our human-driven annot…
Figure 12
Figure 12. Figure 12: In-the-Wild Behavior Framework: Combining density, velocity, and sequentiality of scanpaths as a means to describe gaze-based behaviors. With the combination of these 3 continua, behavior regions, and attributes are defined [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]
Figure 13
Figure 13. Figure 13: Understanding Behavioral Differences Through Distributions: These figures use distributions along different spectra to illustrate the differences in reading behaviors. the feasibility of using advanced deep learning methods for real-time, fine-grained analysis of gaze…
Figure 14
Figure 14. Figure 14: Performance Evaluation of Behavior Classification: Reporting the performance metrics of ML and CNN-based methods on the left table. The 2D CNN achieved the highest Macro F1 score of 0.8. The confusion matrix for the 2D CNN is shown on the right. and in-the-wild behavi…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 26 canonical work pages

  1. [1]

    Seoyoung Ahn, Conor Kelton, Aruna Balasubramanian, and Greg Zelinsky. 2020. Towards Predicting Reading Comprehension From Gaze Behavior. In ACM Symposium on Eye Tracking Research and Applications (Stuttgart, Germany) (ETRA ’20 Short Papers) . Association for Computing Machinery, New York, NY, USA, Article 32, 5 pages. https://doi.org/10.1145/3379156.3391335

  2. [2]

    Kimberly Garcia, Jonas Hermann, Kay Erik Jenß, Yasmine Sheila Antille, and Marc Solèr

    Kenan Bektaş, Jannis Strecker, Simon Mayer, Dr. Kimberly Garcia, Jonas Hermann, Kay Erik Jenß, Yasmine Sheila Antille, and Marc Solèr. 2023. GEAR: Gaze-enabled augmented reality for&nbsp;human&nbsp;activity&nbsp;recognition. In Proceedings of the 2023 Symposium on Eye Tracking Research and Applications (Tubingen, Germany) (ETRA ’23) . Association for Comp...

  3. [3]

    Ralf Biedert, Jörn Hees, Andreas Dengel, and Georg Buscher. 2012. A robust realtime reading-skimming classifier. In Proceedings of the Symposium on Eye Tracking Research and Applications (Santa Barbara, California) (ETRA ’12). Association for Computing Machinery, New York, NY, USA, 123–130. https://doi.org/10.1145/2168556.2168575

  4. [4]

    Paterson, Carsten Schulte, Bonita Sharif, and Sascha Tamm

    Teresa Busjahn, Roman Bednarik, Andrew Begel, Martha Crosby, James H. Paterson, Carsten Schulte, Bonita Sharif, and Sascha Tamm. 2015. Eye Movements in Code Reading: Relaxing the Linear Order. IEEE International Conference on Program Comprehension 2015-Augus (2015), 255–265. https://doi.org/10.1109/ICPC.2015.36

  5. [5]

    Campbell and Paul P

    Christopher S. Campbell and Paul P. Maglio. 2001. A robust algorithm for reading detection. In Proceedings of the 2001 Workshop on Perceptive User Interfaces (Orlando, Florida, USA) (PUI ’01). Association for Computing Machinery, New York, NY, USA, 1–7. https://doi.org/10.1145/971478.971503

  6. [6]

    Megan Caruso, Candace Peacock, Rosy Southwell, Guojing Zhou, and Sidney D’Mello. 2022. Going Deep and Far: Gaze-based Models Predict Multiple Depths of Comprehension During and One Week Following Reading. In Proceedings of the 15th International Conference on Educational Data Mining , Antonija Mitrovic and Nigel Bosch (Eds.). International Educational Dat...

  7. [7]

    Ronald P Carver. 1992. Reading Rate: Theory, Research, and Practical Implications. Source: Journal of Reading 36, 2 (1992), 84–95

  8. [8]

    Xiuge Chen, Namrata Srivastava, Rajiv Jain, Jennifer Healey, and Tilman Dingler. 2023. Characteristics of Deep and Skim Reading on Smartphones vs. Desktop: A Comparative Study. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 516, 1...

Show all 41 references
  1. [9]

    Cole, Karl M

    Zachary J. Cole, Karl M. Kuntzelman, Michael D. Dodd, and Matthew R. Johnson. 2021. Convolutional neural networks can decode eye movement data: A black box approach to predicting task from eye movements. Journal of Vision 21, 7 (2021), 1–15. https://doi.org/10.1167/JOV.21.7.9

  2. [10]

    Eduardo Davalos, Caleb Vatral, Clayton Cohn, Joyce Horn Fonteles, Gautam Biswas, Naveeduddin Mohammed, Madison Lee, and Daniel Levin. 2023. Identifying Gaze Behavior Evolution via Temporal Fully-Weighted Scanpath Graphs. InACM International Conference Proceeding Series . Assoc...

  3. [11]

    Cristina de-la Peña. 2024. Eye-tracking contribution on processing of (implicit) reading comprehension. Journal of New Approaches in Educational Research 13, 1 (12 2024), 0–17. https://doi.org/10.1007/s44322-024-00013-w

  4. [12]

    Steven Elsworth and Stefan Güttel. 2020. Time Series Forecasting Using LSTM Networks: A Symbolic Approach. CoRR abs/2003.05672 (2020), 12 pages. arXiv:2003.05672 https://arxiv.org/abs/2003.05672

  5. [13]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. In2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . IEEE, Las Vegas, NV, USA, 770–778. https://doi.org/10.1109/CVPR.2016.90

  6. [14]

    Timothy Heeren and Ralph D’Agostino. 1987. Robustness of the two independent samples t-test when applied to ordinal scaled data. Statistics in Medicine 6, 1 (1987), 79–90. https://doi.org/10.1002/sim.4780060110 arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1002/sim.4780060110

  7. [15]

    Kevin Hsieh, Amar Phanishayee, Onur Mutlu, and Phillip B. Gibbons. 2020. The non-IID data quagmire of decentralized machine learning. In Proceedings of the 37th International Conference on Machine Learning (ICML’20) . JMLR.org, Article 408, 12 pages

  8. [16]

    Jarodzka and S

    H. Jarodzka and S. Brand-Gruwel. 2017. Tracking the reading eye: towards a model of real-world reading. Journal of Computer Assisted Learning 33, 3 (2017), 193–201. https://doi.org/10.1111/jcal.12189

  9. [17]

    Yu Cin Jian. 2018. Reading instructions influence cognitive processes of illustrated text reading not subject perception: An eye-tracking study. Frontiers in Psychology 9, NOV (11 2018). https://doi.org/10.3389/fpsyg.2018.02263

  10. [18]

    Marcel Adam Just and Patricia A Carpenter. 1980. A Theory of Reading: From Eye Fixations to Comprehension . Technical Report

  11. [19]

    Kaakinen and Jukka Hyönä

    Johanna K. Kaakinen and Jukka Hyönä. 2010. Task Effects on Eye Movements During Reading.Journal of Experimental Psychology: Learning Memory and Cognition 36, 6 (11 2010), 1561–1566. https://doi.org/10.1037/a0020693

  12. [20]

    Takeaki Kariya. 1981. A Robustness Property of Hotelling’s𝑇 2-Test. The Annals of Statistics 9, 1 (1981), 211 – 214. https://doi.org/10.1214/aos/ 1176345350 23 Conference acronym ’XX, June 03–05, 2018, Woodstock, NY Davalos et al

  13. [21]

    Das, Dimitris Samaras, and Gregory Zelinsky

    Conor Kelton, Zijun Wei, Seoyoung Ahn, Aruna Balasubramanian, Samir R. Das, Dimitris Samaras, and Gregory Zelinsky. 2019. Reading Detection in Real-time. In Eye Tracking Research and Applications Symposium (ETRA) . Association for Computing Machinery. https://doi.org/10.1145/3...

  14. [22]

    Peter Kiefer, Ioannis Giannopoulos, and Martin Raubal. 2013. Using eye movements to recognize activities on cartographic maps. In GIS: Proceedings of the ACM International Symposium on Advances in Geographic Information Systems . 478–481. https://doi.org/10.1145/2525314.2525467

  15. [23]

    Hsing Hao Lee, Zih Ling Chen, Su Ling Yeh, Janet Hui Wen Hsiao, and An Yeu Wu. 2021. When eyes wander around: Mind-wandering as revealed by eye movement analysis with hidden markov models. Sensors 21, 22 (11 2021). https://doi.org/10.3390/s21227569

  16. [24]

    Guodong Long, Lu Liu, Tianyi Zhou, Jing Jiang, Wensi Tang, and Michael Blumenstein. [n. d.]. Rethinking 1D-CNN for Time Series Classification: A Stronger Baseline. ([n. d.]). https://doi.org/10.48550/arXiv.2002.10061

  17. [25]

    Xiaochuan Ma, Yikang Liu, Roy Clariana, Chanyuan Gu, and Ping Li. 2023. From eye movements to scanpath networks: A method for studying individual differences in expository text reading. Behavior Research Methods 55, 2 (2 2023), 730–750. https://doi.org/10.3758/s13428-022-01842-3

  18. [26]

    McKnight and Julius Najab

    Patrick E. McKnight and Julius Najab. 2010. Mann-Whitney U Test. John Wiley & Sons, Ltd, 1–1. https://doi.org/10.1002/9780470479216.corpsy0524 arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1002/9780470479216.corpsy0524

  19. [27]

    Mézière, Lili Yu, Genevieve McArthur, Erik D

    Diane C. Mézière, Lili Yu, Genevieve McArthur, Erik D. Reichle, and Titus von der Malsburg. 2024. Scanpath Regularity as an Index of Reading Comprehension. Scientific Studies of Reading 28, 1 (2024), 79–100. https://doi.org/10.1080/10888438.2023.2232063

  20. [28]

    Mézière, Lili Yu, Erik D

    Diane C. Mézière, Lili Yu, Erik D. Reichle, Titus von der Malsburg, and Genevieve McArthur. 2023. Using Eye-Tracking Measures to Predict Reading Comprehension. Reading Research Quarterly 58, 3 (7 2023), 425–449. https://doi.org/10.1002/rrq.498

  21. [29]

    Brian W. Miller. 2015. Using Reading Times and Eye-Movements to Measure Cognitive Engagement. Educational Psychologist 50, 1 (1 2015), 31–42. https://doi.org/10.1080/00461520.2015.1004068

  22. [30]

    Shivsevak Negi and Ritayan Mitra. 2020. Fixation duration and the learning process: an eye tracking study with subtitled videos. Journal of Eye Movement Research 13, 6 (2020), 1–15. https://doi.org/10.16910/jemr.13.6.1

  23. [31]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. 2011. Scikit-learn: Machine Learning in Python. Journal of Machine Le...

  24. [32]

    Ramkumar Rajendran, Daniel T Levin, Anurag Kumar, Gautam Biswas, and Kelly E Carter. [n. d.]. Predicting Learning by Analyzing Eye-Gaze Data of Read-ing Behavior. Technical Report

  25. [33]

    Keith. Rayner. 2012. Psychology of reading. Psychology Press. 486 pages

  26. [34]

    Dewi R. S. Saputro, Alfian F. Hadi, and Gusti N. A. Wibawa. 2022. Profile analysis in clustering with Hotelling’s T-square statis- tics. AIP Conference Proceedings 2566, 1 (11 2022), 040005. https://doi.org/10.1063/5.0116658 arXiv:https://pubs.aip.org/aip/acp/article- pdf/doi/...

  27. [35]

    Rosy Southwell, Julie Gregg, Robert Bixler, and Sidney K. D’Mello. 2020. What Eye Movements Reveal About Later Comprehension of Long Connected Texts. Cognitive Science 44, 10 (10 2020). https://doi.org/10.1111/cogs.12905

  28. [36]

    Namrata Srivastava, Joshua Newn, and Eduardo Velloso. 2018. Combining Low and Mid-Level Gaze Features for Desktop Activity Recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 2, 4 (12 2018), 1–27. https://doi.org/10.1145/3287067

  29. [37]

    Wu Sun, Hui Li, Qingqing Liang, Xiaofeng Zou, Mei Chen, and Yanhao Wang. 2024. On data efficiency of univariate time series anomaly detection models. Journal of Big Data 11, 1 (12 2024). https://doi.org/10.1186/s40537-024-00940-7

  30. [38]

    Oliphant, Matt Haberland, Tyler Reddy, David Cournapeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, Stéfan J

    Pauli Virtanen, Ralf Gommers, Travis E. Oliphant, Matt Haberland, Tyler Reddy, David Cournapeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, Stéfan J. van der Walt, Matthew Brett, Joshua Wilson, K. Jarrod Millman, Nikolay Mayorov, Andrew R. J. Nelson, E...

  31. [39]

    Matthijs J Warrens. 2015. Five ways to look at Cohen’s kappa. Journal of Psychology & Psychotherapy 5 (28 July 2015). https://doi.org/10.4172/2161- 0487.1000197

  32. [40]

    Loschky, and N

    Tianlong Zu, John Hutson, Lester C. Loschky, and N. Sanjay Rebello. 2018. Use of Eye-Tracking Technology to Investigate Cognitive Load Theory. American Association of Physics Teachers (AAPT), 472–475. https://doi.org/10.1119/perc.2017.pr.113

  33. [41]

    † Eth Zurich, Andreas Bulling, Jamie A Ward, Hans Gellersen, and Gerhard Tröster. [n. d.].Eye Movement Analysis for Activity Recognition. Technical Report. Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009 24

Pith tools

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