报告主题：A Recursive Stochastic Algorithm for Real-Time Online Parameter Estimation in Item Response Theory: Enhancing Computational Efficiency for Dynamic Educational Assessment

报 告 人：陆 婧 副教授

报告时间：2024年12月7日（周六）下午 16：00-18：00

报告地点：腾讯会议（会议号：644-540-794）

报告摘要: Traditional large-scale educational data is often static and updated periodically, making it difficult to reflect the dynamic changes in the educational environment in real time. With advancements in technology, many online exam platforms and educational assessment tools (such as adaptive tests) can collect students’ response data and behavioral data in real time during the examination process. Most estimation methods in Item Response Theory (IRT), while highly accurate, are mostly used for parameter estimation in offline environments. When real-time data continuously arrives and online parameter estimation is needed, these methods become computationally impractical. To address this issue, this paper proposes a recursive stochastic algorithm--average truncated stochastic Newton algorithm--for the quick online estimation of parameters in the IRT model. The proposed algorithm can also provide parameter estimates in offline environments and demonstrates significant improvements in computational efficiency compared to the EM algorithm in the mirt package in R software. In situations requiring faster approximation methods, our proposed approach can serve as a powerful alternative to the offline EM method. Furthermore, we investigate the asymptotic properties of the algorithm, proving its almost sure convergence and asymptotic normality. Numerical experiments on both simulated and real data also demonstrate the practicality of the proposed method.

报告人简介：陆婧，东北师范大学数学与统计学院副教授、博士生导师。研究方向为教育统计与心理测量。主持并完成国家自然科学基金青年项目一项，中国博士后科学基金特别资助站中项目及中国博士后科学基金面上项目各一项。在Psychometrika、Multivariate Behavioral Research、British Journal of Mathematical and Statistical Psychology、Journal of Mathematical Psychology、Journal of Educational and Behavioral Statistics、Journal of Educational Measurement、Frontiers in Psychology等学术期刊发表论文多篇。目前研究兴趣主要有作答反应时间的建模及其应用、作答反应行为分析、缺失数据分析、测量模型的相关算法开发、认知诊断模型等。