Signals Before The Boom : Identifying Emerging Content Formats on Social Video Platforms

Sammanfattning

In the black box methodology of predictive modeling, using metadata on social video platforms alongside past performance to predict future success is nothing new. This study attempts to build additional layering on top of this reasoning method by using natural language processing to capture the style and character of videos, making visible the underlying reasons why a specific video format outperforms another. This approach aims to transform an ambiguous model into an interpretable one, allowing for more informed decision-making. The main contribution of this study is the Signals Before the Boom (SBB) framework, a modular pipeline spanning from data collection to a user interface. The design is intentionally flexible, allowing modules to be exchanged to process alternative inputs or generate different outputs. The proof-of-concept model presented in this paper was developed through an iterative process rooted in Design Science Research, utilizing short sprints oscillating between development and evaluation. To ensure practical relevance on the commercial side of digital media, the quantitative work was supplemented by two expert interviews. Methodologically, the framework utilizes an API to retrieve videos and transcripts, which are stored in a database. Large Language Models (LLMs) and an S-BERT model are then used to classify and convert the textual data into numerical, semantically encoded vectors. These vectors are grouped into clusters of structurally similar videos, evaluated, and ranked using an engagement-based emergence score. Finally, the top-performing clusters are analyzed by an LLM to decode the underlying structural reasons for their success. Applied to a dataset of YouTube videos from April-May 2026, the SBB framework successfully mapped the content landscape into 47 distinct semantic clusters. By identifying formats with accelerating momentum, the study demonstrates that weak signals can be surfaced early. The final output is presented through an Emerging Content Format Dashboard, bridging the gap between statistical trends and actionable insights for content creators.