Article

Temporal Topic Modeling of Netflix Descriptions Using TF-IDF and NMF to Map the Evolution of Digital Storytelling Themes

---

Abstract

This study investigates the thematic evolution of storytelling within Netflix’s global catalog between 1990 and 2021 through a computational text analysis framework. Using TF-IDF vectorization and Non-Negative Matrix Factorization (NMF), the research identifies latent narrative patterns from plot descriptions and traces their temporal trajectories to map the digital zeitgeist of streaming media. The dataset, derived from Netflix’s public metadata, was preprocessed to remove noise and standardized through tokenization and lemmatization. Vocabulary pruning produced a corpus of 2,516 unique terms, which served as the input for topic modeling. After evaluating model coherence and interpretability, a ten-topic configuration was selected as optimal. The extracted themes encompassed crime and action, adventure and fantasy, documentary and biography, romance and youth, family life, and comedy. Temporal aggregation and visualization revealed a clear diversification of thematic focus over three decades: early periods were dominated by crime and domestic dramas, while later years showed the rise of documentary and fantasy narratives alongside enduring romantic motifs. The findings highlight how Netflix’s content strategy mirrors broader cultural shifts, balancing realism, escapism, and emotional intimacy. Although limited by its focus on short textual summaries, the study demonstrates the capability of unsupervised topic modeling to capture evolving storytelling trends and provides an empirical basis for understanding how global streaming narratives reflect and shape cultural consciousness in the digital age.

Keywords: Netflix, Topic Modeling, TF-IDF, Narrative Evolution, Digital Zeitgeist

How to Cite:

Wahyuningsih, T. & Hananto, A., (2025) “Temporal Topic Modeling of Netflix Descriptions Using TF-IDF and NMF to Map the Evolution of Digital Storytelling Themes”, Journal of Digital Society 1(4), 287-298. doi: https://doi.org/10.63913/jds.v1i4.43