I’m so proud of my students, Sohaila Eltanbouly (prospective PhD student) and Salam Albatarni (MSc student), whose work established a new state-of-the-art performance (i.e., outperforming the best-ever published solution to the problem on a widely-used dataset) in Automated Essay Scoring (AES) by introducing a novel framework, called TRATES. TRATES full paper has just been accepted at ACL 2025 Findings, concluding more than 18 months of their hard work to get it published at the top NLP conference!
The main idea of TRATES is simple but elegant; instead of asking LLMs to directly score the written essays, ask them to suggest features and to extract the values of such features, then train a traditional regression model on those (and other generic) features!
TRATES is a pure work of our research group at Qatar University, and funded by Qatar Research, Development and Innovation (QRDI) Council.
This work is dedicated to our people in Gaza! Your perseverance is a continuous motivation for us!
The final version of the paper is available here.
Leave a Reply