Sequence Aware Representation Learning Traces Evolving Teaching Practices Through Historical Curriculum Records

Abstract

The continuous process of curriculum change plays an important role in making sure that the curriculum stays of high quality and promotes competency-based learning while also taking into consideration the needs of the ever-changing workforce. Traditional methods of curriculum analysis usually consider each version of the curriculum separately and are unable to incorporate the relation of curriculum versions to changes over time. This limitation hinders their ability to identify any changes in the way the instruction or material delivery has been changed. In order to solve this problem, in this paper, a novel Sequence-Aware Representation Learning (SARL) model for tracking the evolution of teaching practices using historical curriculum data of higher education institutions in Uzbekistan is introduced. The proposed SARL model considers curriculum revisions as a sequence of events and connects them to several aspects of education, such as the structure of the course, learning outcome, assessment, relationship of prerequisites, and credits, using the unified feature representation. The use of learned representations will help analyze the way the curriculum is being changed and detect any changes in the instruction as a valuable basis for decision-making regarding curriculum planning and educational policies. The framework has been verified using the curriculum history dataset collected from the universities of Uzbekistan, in comparison with such approaches as Logistic Regression, Random Forest, LSTM, and Bi-LSTM. According to the results of the experiments, the proposed SARL framework proved to be effective with the following metrics in terms of accuracy (96.84 %), precision (96.41 %), recall (96.12 %), F1 score (96.26 %), and AUC (97.31 %). Furthermore, the framework detected 96.85% of curriculum changes, 96.43% of teaching practices, 97.18% of sequence consistencies, and 96.79% of curriculum evolution predictions, demonstrating the effectiveness in capturing long-term trends of instruction changes.