Video Demo: https://www.youtube.com/watch?v=B5DKbSFF97g
This web app generates a report for a Spotify playlist that includes: lyrics sentiment score, lyrics lexical diversity, audio valence and audio energy, among others.
- Database: SQLite
- Back-end: Python (Flask)
- Front-end: Javascript, Bootstrap
The web app is Python based its dependencies are managed with Poetry. To install the dependencies ensure you have Python 3.10 and Poetry installed in your machine. Then, execute the following command from the parent directory
poetry install
The web app can then be run locally by executing the following commands from the parent directory:
poetry run flask --app playlist_sentiment_analyser init-db
poetry run flask run
The app can be divided in three big parts: data pipeline (ETL), data analysis and data visualization.
The app gets a playlist's track information from Spotify using the Spotipy library, which is a convenient python interface for the official Spotify Web API. This information includes track name, artists, album and audio features.
Lyrics for each track are extracted from Genius.com using the LyricsGenius library, which is a Python client for the Genius.com API that additionally scraps a track's lyrics.
The lyrics data is then cleaned and loaded to a SQLite database.
For simplicity only lyrics in English are analysed. The app uses one of Fasttext's language identification models to label the lyrics language. It subsequently performs sentiment analysis on the remaining lyrics, leveraging the Natural Language Processing library Spacy.
The data is visualized via a very simple dashboard designed with Bootstrap. Figures are mainly done using plotly which produces beautiful plots that can be easily embedded into an html template, check this blog post for more detail on how to do so. Finally, the word cloud was produced with the word_cloud library and embbeded into the html using the trick described here.