Inspiration
With the advent of short-form media and other new ways to share music, many new artists have gotten the chance to be in the spotlight. We were curious—are there patterns in trending songs, and can budding musicians leverage these patterns to become the next superstar?
What it does
Hit the Charts! lets musicians input their song lyrics and provides an analysis of their themes, sentiment, and similarity to past Billboard chart-toppers. It estimates the song’s potential popularity and highlights the most thematically similar hit songs, helping users get insights into what makes a hit.
We also provide AI-driven recommendations on how to improve song lyrics to boost their potential popularity. Additionally, users can explore the Songs Collection page, where they can filter charting songs by month and year and discover how songs cluster together based on themes and sentiment trends. On the Playground page, users can experiment with different themes, keywords, and sentiment to see how they impact predicted popularity.
How we built it
- Database Creation: We collected data on Billboard charting songs from the past three years, including chart history and Spotify’s popularity scores.
- Theme Analysis: Using OpenAI’s Llama LLM, we extracted high-level themes (e.g., love, heartbreak, regret) and performed sentiment analysis using VADER.
- Lyric Matching: We encoded user-submitted lyrics and compared them to our song database using feature similarity. This allowed us to predict popularity and recommend the most thematically similar charting songs.
- Interactive Features: We built filtering and visualization tools to explore trends in song themes and sentiment, along with a playground for testing custom lyrics and combinations of themes.
Challenges we ran into
- Data Collection: Aggregating reliable and comprehensive data on charting songs was time-consuming and required filtering incomplete or inconsistent information.
- Theme Extraction: Fine-tuning OpenAI’s LLM to generate meaningful and consistent themes was a challenge, especially for abstract or metaphorical lyrics.
- Interactive Visualizations: Creating dynamic filters and real-time popularity predictions on the playground page while ensuring the interface remained intuitive was complex.
Accomplishments that we're proud of
- Building a robust database of charting songs and integrating multiple analysis tools (LLMs, VADER, and Spotify data).
- Successfully creating an intuitive and accurate similarity-matching system for song lyrics.
- Designing an interactive playground where users can tweak their lyrics and see the impact on predicted popularity.
- Providing valuable insights to musicians about what makes their song comparable to existing hits.
What we learned
- The power of combining multiple NLP tools for a comprehensive analysis of text data.
- How to build scalable song-matching algorithms using feature similarity and embeddings.
- The importance of high-quality data in generating meaningful predictions and insights.
- Designing interactive experiences that make complex AI tools user-friendly and fun.
What's next for Hit the Charts!
- Expanding our dataset and releasing it publicly so others can benefit and perform their own analysis.
- Improving our AI recommendations to give even more personalized feedback to musicians.
- Enhancing the playground with more customization options and real-time theme visualization to help users craft the perfect song.
Log in or sign up for Devpost to join the conversation.