FTI UKDW Lecturer Develops Essay Scoring System Based on Text Similarity
The role of Learning Management Systems (LMS) has grown significantly in the post-COVID-19 era, as educators and students have become increasingly accustomed to online and blended learning models. This shift has also led to a growing demand for online assessments and automated grading systems.
While most LMS platforms already offer robust tools for grading multiple-choice and true/false questions, automatically scoring open-ended responses remains a substantial challenge. This gap has highlighted the need for improved Automated Short Answer Grading (ASAG) systems.
Addressing this need, a research team from the Informatics Program at Duta Wacana Christian University (UKDW) Yogyakarta is developing a new ASAG system based on text similarity detection. The team—comprising Dr. Phil. Lucia Dwi Krisnawati, Aditya Wikan Mahastama, S.Kom., M.Cs., Natanael Tegar Pramudya, and Excelsior Valentino Yonathan—aims to create a system capable of assessing descriptive and narrative responses by evaluating their semantic and lexical resemblance to model answers.
Dr. Lucia, the lead researcher, explained that the project builds upon previous work on text similarity detection, with the goal of adapting that research into a practical application for automated essay scoring. The team also plans to integrate the system into existing LMS platforms, including UKDW’s own eClass system.
The architecture of the system consists of two main modules: a text similarity detection module and a scoring module. The first module calculates the semantic and lexical similarity between a student’s answer and the reference answer provided by the teacher, generating vector scores for comparison.
The research follows an experimental quantitative methodology. “For data collection, we used exam questions and assignments from courses we teach. We also gathered training data from Stella Maris Junior High School in Surabaya. The use case for this system is based on real-world scenarios, where teachers or lecturers create questions and provide answer keys,” said Dr. Lucia.
The system evaluates responses using a combination of semantic and lexical similarity vectors, answer length, and vocabulary richness. These features are then processed using machine learning models to predict scores through classification and regression techniques.
The research has produced a functional and scalable prototype that can be embedded within LMS environments. The prototype has been officially registered and granted intellectual property rights.
Integrating text similarity detection into essay scoring systems significantly reduces the grading burden on teachers and lecturers. It also encourages the use of open-ended questions in LMS-based assessments, promoting deeper learning and helping students hone their writing skills.
Credits: W-ID & Doc: humasukdw | Trans. EN: drr
