EUVIMEDEUVIMED entdecken
Crossref

Attention-based automated essay scoring in educational assessment systems over wireless network platforms

Sarvinoz Otamuratova, Natalya Kim, Dilfuzakhon Kozokboeva, Nozima Khamdamova, Nazokat Panjieva, Minavvarxon Yunusova, Nafisa Hamroyeva, Muhayyo Roziyeva

Interpersona An International Journal on Personal Relationships · 17.11.2026 · 20 (4) · e2365

DOI: 10.54899/ijpr.v20i4.2365 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Interpersona An International Journal on Personal Relationships, online erschienen am 17.11.2026. Verfasst von Sarvinoz Otamuratova und 7 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Existing automated essay scoring (AES) models are either opaque to teachers or too demanding of mobile and wireless infrastructures that many students use to turn in their work, and manual essay grading is time-consuming and expensive while also subject to rater drift. In this work, both problems are tackled, and a hierarchical trait-aware attention network (HiTA-AES) is adopted, which is deployed in an edge-assisted wireless assessment platform. The model uses bidirectional gated recurrent units to encode essays at word and sentence levels, as well as additive attention at each level, and four trait queries aligned with the rubric for content, organization, language, and conventions. Linguistic features are integrated by a learned gate, while mean-squared error, pairwise ranking, and knowledge distillation from a language-trained transformer teacher are used in training. This model was quantized to 8-bit precision and deployed on an edge node that is accessible using Wi-Fi 5, Wi-Fi 6, 4G LTE, and 5G access. Five-fold cross-validation was performed for experiments on the eight prompts of the public Automated Student Assessment Prize corpus. HiTA-AES attained an average quadratic weighted kappa of 0.798 (95% CI: 0.789–0.807), outperforming a fine-tuned BERT model (with 110 million parameters) by 0.017 (Wilcoxon p = 0.003) with its 3.1 million parameters. Pearson correlation was 0.838 and adjacent agreement was 93.6%. On Wi-Fi 6, mean end-to-end scoring latency was 96ms with 100 concurrent users, and with an adaptive store-and-forward protocol, 98.8% submission success was maintained with 10% packet loss. The results of ablation indicated that word-level attention played the largest role in the scoring accuracy. Results show that interpretable, lightweight attention-scoring can provide reliable, near real-time, formative feedback in a wireless classroom with a diversity of devices.

Zitieren

APA

Otamuratova, S., Kim, N., Kozokboeva, D., Khamdamova, N., Panjieva, N., Yunusova, M., Hamroyeva, N., & Roziyeva, M. (2026). Attention-based automated essay scoring in educational assessment systems over wireless network platforms. Interpersona An International Journal on Personal Relationships, 20(4), e2365. https://doi.org/10.54899/ijpr.v20i4.2365

Vancouver

Otamuratova S, Kim N, Kozokboeva D, Khamdamova N, Panjieva N, Yunusova M, et al. Attention-based automated essay scoring in educational assessment systems over wireless network platforms. Interpersona An International Journal on Personal Relationships. 2026;20(4):e2365. doi:10.54899/ijpr.v20i4.2365

BibTeX
@article{Otamuratova2026,
  author = {Otamuratova, Sarvinoz and Kim, Natalya and Kozokboeva, Dilfuzakhon and Khamdamova, Nozima and Panjieva, Nazokat and Yunusova, Minavvarxon and Hamroyeva, Nafisa and Roziyeva, Muhayyo},
  title = {Attention-based automated essay scoring in educational assessment systems over wireless network platforms},
  journal = {Interpersona An International Journal on Personal Relationships},
  year = {2026},
  volume = {20},
  number = {4},
  pages = {e2365},
  doi = {10.54899/ijpr.v20i4.2365}
}
Kostenlos starten →