Evaluation of Deep Clustering for Assessing Undergraduate Understanding in Ideological and Political Education: Data-Driven Analytics

Assessing undergraduate students’ understanding of ideological and political education is vital for effective teaching and curriculum improvement. Traditional evaluation methods often need to be improved in capturing the complexity and depth of students’ comprehension. This paper explores deep clustering algorithms as a data-driven analytics approach for evaluating undergraduate ideological and political education understanding. By leveraging the power of deep learning and clustering techniques, this research seeks to provide objective and insightful assessments to enhance instructional strategies and improve student outcomes. This paper proposes a deep document clustering via a multi-layer subspace semantic fusion model. First, the model uses a deep autoencoder to extract the potential semantic representation of text at different levels. Then, a multi-layer subspace semantic fusion strategy is designed to map the semantic representation of different layers to different subspaces to obtain fusion semantics and use it for clustering. Additionally, a joint loss function is designed by using the self-representation loss of subspace clustering to monitor the updating of model parameters. The experimental results demonstrate that high clustering precision, normalized mutual information, and adjusted Rand index in the assessment of undergraduate understanding in ideological and political education signify the effectiveness of the clustering approach in accurately grouping students based on their comprehension levels. These metrics provide valuable information to educators, allowing them to gain insights into students’ understanding, identify patterns or clusters of students with similar comprehension, and tailor educational interventions or strategies to meet individual or group needs.