We introduce a predictive model designed to identify potential hit songs for short video advertising. This model could offer valuable insights for advertisers seeking to stay ahead of music trends and align their content accordingly.
- Classification
- Regression
- Similarity Matching
The Random Forest Classifier performed the best, after which Cosine Similarity Matching of top 10 similar records also performed well. These have been applied on real-time data collected from Spotify's current top chart playlists and can be applied to today's demographics taste. The spotipy library and Spotify API was used for data collection.
The report proposes a strategy for TikTok, suggesting the curation of a list of songs anticipated to become popular. This curated list can simplify content creation for TikTok and its creators, leading to increased user engagement and potential ad revenue growth. To validate the model's effectiveness, we tested it on a list of new songs added to Spotify, resulting in the identification of the top three songs: SAY MY GRACE by Offset ft. Travis Scott, Hope You Know by Kodak Black, My Simple Jeep by Eyedress (feat. Mac DeMarco).
By allocating resources to develop templates and proactively promoting these songs to advertisers, TikTok can position itself as a frontrunner in capitalizing on music trends. This repository includes our models and data collection efforts, which have been instrumental in shaping this strategic approach for TikTok's growth and profitability.
For a more detailed understanding of the project, please refer to the full document.
Access the full report here.
View the presentation deck here.
This project was created for Duke Fuqua School of Business: DECISION 520Q - Data Science of Business
- Roshni Balasubramanian
- Dhruv Arora
- Camilla Kang
- Haider Ali
Team Members: