همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

LLU-460: A Dynamic Dataset of Large Language Model Usage Among Students at Islamic Azad University, Karaj

نویسندگان :

Alireza Jafari¹ Seyed Mohammad Amin Alemohammad² Niloofar Mirzaei Chahardeh³

1- dept. of Computer Engineering, KA.C., Islamic Azad University Karaj, Iran 2- dept. of Computer Engineering and Sciences, Islamic Azad University, Central Tehran Branch 3- dept. of Computer Engineering, SR.C., Islamic Azad University Tehran, Iran

کلمات کلیدی :

llms،web application،llu-460،xgboost،random forest

چکیده :

Abstract— In recent years, the use of large language models and AI tools has become increasingly common among university students, yet their actual usage patterns remain poorly understood. To investigate this, we developed an interactive web application that collected responses from students at Islamic Azad University, Karaj. Unlike traditional surveys, the system adapted questions based on each student’s prior answers, enabling more precise and context-aware data collection. Using this platform, 460 complete responses were gathered. Although the dataset is moderate in size, it reveals meaningful patterns, including the frequency of AI tool usage, preferred tools, and students’ perceptions of their academic utility. A notable aspect is the slightly higher proportion of male participants, introducing a minor imbalance. Despite this, the dataset reflects real student behavior and can serve as a valuable resource for future research. For modeling and prediction, we applied two supervised learning algorithms: Random Forest and XGBoost. XGBoost performed consistently on this limited dataset, achieving approximately 0.80 accuracy, whereas Random Forest showed signs of overfitting with unrealistically perfect results. Overall, our findings demonstrate that real student data can model AI tool usage patterns and suggest that expanding the dataset will enhance the reliability of future analyses. The dataset and web application are publicly available on Zenodo [Online].