Data-driven Podcast Advertising: A Novel Framework

Abstract

The growing podcast industry necessitates innovative advertising strategies, prompting this study to explore the use of data science methods to enable native advertisement placement, in which ads align with the surrounding content. This research seeks to locate advertisement spots at points of topical shifts as well as assign meaningful topics to the content surrounding those shifts in podcast transcripts. A transformer-based clustering approach, integrated with a text segmentation algorithm, is developed for this purpose, advancing previous literature in text segmentation. By modelling the Spotify Podcast Dataset, the developed methodology’s ability to identify meaningful advertisement spots in podcasts and assign topics from the corpus to these segments is validated. This proof-of- concept study not only technically enables native advertising but also proposes a business framework for its monetization, outlining potential integration into podcast platforms. The study also positions the relevance of the methodology in relation to network effects and platform theory.