Multimodal Clustering of Students Using Social Media Profiles to Analyze Academic Performance

Sergei Gorshkov

Keywords: multimodal clustering, multimodal learning analytics, social media analysis, machine learning in education, digital footprint

Abstract: This study explores the relationship between students' digital behavior and academic performance using a multimodal clustering framework applied to social media profiles. Unlike traditional self-reported surveys, it objectively analyzes students’ social media footprints by integrating text and image embeddings through state-of-the-art machine learning models, including Sentence-BERT, CLIP, BERTopic, HDBSCAN, and KMeans++ clustering. Using a dataset of 2,909 students, the study identifies 52 distinct behavioral clusters, revealing that engagement in educational and scientific content is associated with higher academic performance, while entertainment-focused digital habits are associated with lower GPAs. Statistical analyses confirm significant differences across clusters, highlighting structured associations between digital behavior patterns and academic outcomes. These findings contribute to AI-driven education analytics and demonstrate how multimodal machine learning and big data analytics can support large-scale analysis of student digital behavior in relation to academic performance. Future research should explore longitudinal trends to refine the interpretation of these patterns and further extend intelligent learning analytics.

Submission Number: 141