CDLM-AA

CDLM-AA: Cognitive Diagnostic Language Model for Adaptive English Writing Assessment Overview CDLM-AA is an intelligent cognitive diagnosis and adaptive assessment system for English writing ability, integrating pre-trained language models (e.g., BERT or GPT) with cognitive diagnostic theory. The framework overcomes the limitations of traditional rule-based and human-evaluated writing assessments by providing scalable, context-sensitive, and personalized feedback. At its core, CDLM-AA introduces: CDLM (Cognitive Diagnostic Language Model) — a multimodal, transformer-based language architecture aligned with cognitive attributes. Adaptive Assessment Mechanism — dynamically adjusts task difficulty based on individual learner profiles. Innovative Methodologies — graphical propagation layers and spatial-temporal attention for fine-grained diagnostic evaluation. ✨ Features 1. Cognitive Diagnostic Language Model (CDLM) Built on pre-trained language models (e.g., BERT, GPT), fine-tuned on annotated writing samples. Maps writing samples to latent cognitive attributes (grammar, coherence, vocabulary, organization). Multimodal Encoder + Encoder–Decoder architecture integrates textual and auxiliary inputs. Sparse Mixture-of-Experts dynamically routes representations for specialized cognitive tasks.📎 See Figure 1 on page 7 for the overall architecture diagram of CDLM. 2. Multimodal Encoder Architecture Employs static and dynamic channel mixing for robust feature refinement. Balances adaptability with stable structure to support different writing genres.📎 Figure 2 on page 8 shows the dual-path encoding structure with static/dynamic attention blocks. 3. Graphical Propagation & Attention Mechanisms Graphical Propagation Layer dynamically adjusts weighting of attributes according to learner performance. Spatial-Temporal Attention focuses on critical textual segments, highlighting grammar, coherence, and structure. Real-time iterative recalibration aligns feedback with learner progress.📎 Figures 3–4 (pages 10–11) illustrate the feature mapping and graph-based attention process. 4. Adaptive Assessment Selects tasks using a utility function targeting weak attributes. Task difficulty and focus evolve with learner proficiency, ensuring continuous engagement and growth. 📊 Datasets Dataset Description Purpose English Writing Skill Evaluation Dataset Writing samples annotated for grammar, vocabulary, coherence Model training and benchmarking Cognitive Writing Assessment Dataset Think-aloud protocols, keystroke logs, cognitive writing behaviors Cognitive process modeling Language Model Guided Writing Dataset Human-model interactive writing tasks LLM-assisted feedback testing Adaptive English Writing Proficiency Dataset Adaptive tasks adjusted by performance Real-time feedback evaluation 🚀 Usage Attribute mastery scores (grammar, coherence, vocabulary, organization) Attention visualization on critical text segments Personalized writing feedback Adaptive prompt selection for next tasks 🧪 Applications Automated, adaptive English writing evaluation Cognitive profiling for targeted learning interventions Real-time formative feedback in digital learning environments Scalable assessment for large educational settings 🧩 Model Components Component Description CDLM Transformer-based cognitive diagnostic language model Multimodal Encoder Static + dynamic channel mixing for feature integration Graphical Propagation Layer Adjusts task focus dynamically Spatial-Temporal Attention Highlights key textual elements for feedback Adaptive Feedback Loop Iteratively updates learner profile and prompt difficulty 📈 Performance Dataset Accuracy Precision Recall F1 Score English Writing Skill Evaluation 89.78 89.23 88.56 88.89 Cognitive Writing Assessment 90.67 90.12 89.56 89.89 Language Model Guided Writing 89.12 88.56 87.90 88.23 Adaptive Writing Proficiency 90.03 89.60 88.94 89.27 📎 Tables 1–2 on pages 15–16 compare CDLM-AA to ResNet, ViT, I3D, BLIP, DenseNet, and MobileNet, showing consistent gains across all metrics. 🧭 Future Work Expand cross-linguistic support for multilingual writing assessment. Enhance efficiency via lightweight model variants for classroom deployment. Integrate explainable AI for transparent feedback interpretation. Enable real-time adaptive learning loops for live educational applications. 📜 License This project is licensed under the MIT License. 🙏 Acknowledgments This work was supported by the Higher Education Research Program of Hainan Province (Grant No. Hnjg2023-58).Authors: Minghui Chen, Yang Tian, Shenggu Chen, Zhengzheng Huang.This research bridges cognitive diagnostic theory with pre-trained language modeling, advancing adaptive and scalable English writing assessment.