Scholarlink

<p>Traditional manual essay grading methods struggle to evaluate student writing quickly and impartially. Automated essay scoring (AES) systems use NLP to assess text, yet overlook behavioral context and lack fairness in real-time feedback. The research presents DeepGrade, an IoT-enhanced automated essay scoring system for context-aware English essay grading based on deep learning. Score accuracy and bias can be enhanced by combining semantic text interpretation with real-time behavioral information, such as keystroke dynamics and writing behaviors. A transformer-based language model (BERT) encodes essay content, and LSTM-based spatiotemporal encoders handle IoT-derived data in the hybrid method. These two feature streams are combined and scored by a fairness-aware scoring brain to produce holistic, rubric-aligned scores. In experiments using the ASAP 2.0 dataset and synthetically produced IoT-like behavioral traits to simulate writing interactions, the researchers achieve a QWK of 0.89. Though intriguing, these results are proof-of-concept and need real-world behavioral data validation for adoption. Score bias across demographic categories is reduced by 9.6% using the model. Adding IoT behavior boosts rubric-level scoring reliability and prediction precision by 12%. Economically disadvantaged groups’ score variance drops 15.4% due to a reduction in fairness awareness. For accurate, fair, and real-time essay assessment in smart educational contexts, DeepGrade: IoT is scalable and intelligent. This conceptual prototype shows feasibility, but full-scale deployment requires greater validation in real-world educational settings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An automated english essay scoring system based on deep learning and the internet of things

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月25日

查看全文
  • Wu Tiantian

摘要

Traditional manual essay grading methods struggle to evaluate student writing quickly and impartially. Automated essay scoring (AES) systems use NLP to assess text, yet overlook behavioral context and lack fairness in real-time feedback. The research presents DeepGrade, an IoT-enhanced automated essay scoring system for context-aware English essay grading based on deep learning. Score accuracy and bias can be enhanced by combining semantic text interpretation with real-time behavioral information, such as keystroke dynamics and writing behaviors. A transformer-based language model (BERT) encodes essay content, and LSTM-based spatiotemporal encoders handle IoT-derived data in the hybrid method. These two feature streams are combined and scored by a fairness-aware scoring brain to produce holistic, rubric-aligned scores. In experiments using the ASAP 2.0 dataset and synthetically produced IoT-like behavioral traits to simulate writing interactions, the researchers achieve a QWK of 0.89. Though intriguing, these results are proof-of-concept and need real-world behavioral data validation for adoption. Score bias across demographic categories is reduced by 9.6% using the model. Adding IoT behavior boosts rubric-level scoring reliability and prediction precision by 12%. Economically disadvantaged groups’ score variance drops 15.4% due to a reduction in fairness awareness. For accurate, fair, and real-time essay assessment in smart educational contexts, DeepGrade: IoT is scalable and intelligent. This conceptual prototype shows feasibility, but full-scale deployment requires greater validation in real-world educational settings.