Series/Report no.:
ITS 33rd European Conference 2025: "Digital innovation and transformation in uncertain times", Edinburgh, UK, 29th June – 1st July 2025
Abstract:
This study examines contextual patterns in over-the-top (OTT) streaming consumption through deep learning-driven cluster analysis, advancing our understanding of contemporary media repertoires. Drawing on media repertoire theory, we analyze how users develop personalized combinations of media platforms and usage patterns across temporal, spatial, and technological contexts. Using longitudinal diary data from the Korea Media Panel (2021-2024), we apply Gated Recurrent Unit (GRU) autoencoders to compress 12-day behavioral sequences into latent representations, subsequently identifying distinct consumption clusters through k-means analysis. Our findings reveal three primary usage repertoires: Conventional Home Viewing (51%), characterized by traditional single-device evening consumption; Heavy Prime-Time (29%), featuring intensive multi-device usage during evening hours; and Daytime Flexible (20%), demonstrating distributed consumption throughout the day across mobile devices. Multinomial regression analysis confirms significant demographic and behavioral differences between clusters, with age, gender, income, and OTT usage frequency serving as key predictors of cluster membership. The study contributes methodologically by demonstrating how sequence-based deep learning can operationalize media repertoire theory without imposing restrictive analytical assumptions. Practically, our findings illuminate strategic opportunities for telecommunications operators and content providers in mature OTT markets. The identification of distinct contextual usage patterns—from concentrated prime-time multi-screening to distributed mobile consumption—suggests targeted service design strategies. Furthermore, the relationship between consumption repertoires and mobile carrier choice reveals how infrastructure providers must consider usage patterns beyond aggregate metrics. This research establishes a computational framework for analyzing technology-mediated behavioral patterns in high-connectivity environments.