Technology Category
Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.Machine Learning: Deep Generative Models & AutoencodersConstraint Satisfaction and Optimization: Satisfiability Modulo Theories
TransGAT: Transformer-Based Graph Neural Networks for Multi-Dimensional Automated Essay Scoring
🤖 AI Summary
This work addresses two key limitations in automated essay scoring (AES): insufficient semantic modeling—particularly ambiguity arising from polysemy—and the neglect of local writing features such as grammar, lexical choice, and coherence. To this end, we propose TransGAT, the first model to integrate fine-tuned Transformer encoders (BERT, RoBERTa, or DeBERTaV3) with a Graph Attention Network (GAT). TransGAT explicitly captures structural dependencies among words by constructing dependency parse graphs and employs a dual-stream architecture to jointly encode sentence-level contextual representations and token-level graph-structured representations. This enables fine-grained, multi-dimensional scoring. Evaluated on the ELLIPSE dataset, TransGAT achieves an average quadratic weighted kappa of 0.854—significantly outperforming existing baselines—and enhances both scoring consistency and interpretability.
Technology Category
Application Category
📝 Abstract
Essay writing is a critical component of student assessment, yet manual scoring is labor-intensive and inconsistent. Automated Essay Scoring (AES) offers a promising alternative, but current approaches face limitations. Recent studies have incorporated Graph Neural Networks (GNNs) into AES using static word embeddings that fail to capture contextual meaning, especially for polysemous words. Additionally, many methods rely on holistic scoring, overlooking specific writing aspects such as grammar, vocabulary, and cohesion. To address these challenges, this study proposes TransGAT, a novel approach that integrates fine-tuned Transformer models with GNNs for analytic scoring. TransGAT combines the contextual understanding of Transformers with the relational modeling strength of Graph Attention Networks (GAT). It performs two-stream predictions by pairing each fine-tuned Transformer (BERT, RoBERTa, and DeBERTaV3) with a separate GAT. In each pair, the first stream generates essay-level predictions, while the second applies GAT to Transformer token embeddings, with edges constructed from syntactic dependencies. The model then fuses predictions from both streams to produce the final analytic score. Experiments on the ELLIPSE dataset show that TransGAT outperforms baseline models, achieving an average Quadratic Weighted Kappa (QWK) of 0.854 across all analytic scoring dimensions. These findings highlight the potential of TransGAT to advance AES systems.
Problem
Research questions and friction points this paper is trying to address.
Automated essay scoring lacks contextual understanding of polysemous words
Existing methods overlook specific writing aspects like grammar and cohesion
Current GNN approaches use static embeddings instead of contextual ones
Innovation
Methods, ideas, or system contributions that make the work stand out.
Transformer-GNN fusion for contextual essay scoring
Two-stream prediction with syntactic dependency graphs
Analytic scoring across multiple writing dimensions
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
King Abdulaziz University
