Classifying song lyrics using Natural Language Processing (NLP)

In this project, we build a text classification model on song lyrics. The task is to predict the artist from a piece of text. To train such a model, we first need to collect a lyrics dataset. We will

• Download a HTML page from lyrics.com with links to songs using the requests library
• Extract hyperlinks of song pages using the BeautifulSoup library
• Download and extract the song lyrics and save them to a temporary CSV file using the requests and pandas libraries
• Clean and preprocess the lyrics using TreebankWordTokenizer and WordNetLemmatizer from the nltk library
• Vectorize the text using TfidfVectorizer from the sklearn library
• Build a classification model using Naive Bayes classifier for multinomial models (MultinomialNB) and tune its hyperparameters using GridSearchCV
• Predict the artist from a piece of text based on the trained model

All these steps are implemented in the files contained in includes. To run the project, create a Python environment (Python 3.11), install dependencies from requirements.txt, define configuration in settings.py, and run main.py in the root directory. To predict the artist from a piece of text, run predict.py in the root directory.

Running main.py with all options set tu True will create the following files in the data and models directories:

• data/songs_clean.csv will contain the lyrics of ~600 songs from 3 artists (Adele, Eels, Rage Against The Machine)
• data/songs_by_line.csv will contain the same lyrics split by line (~15.000 rows)
• models/trained_model.pkl will contain the trained model

The trained model is included in the project. To just try out the prediction, you can run predict.py without running main.py first.

The Jupyter Notebook included in the proyect uses the the functions defined in the files in include to walk through the steps of the script.

The script includes the possibility to create word clouds from the corpus. See the function plot_wordcloud() in includes/misc.py or the example the Jupyter Notebook. To create the text option, download the Boldova font first and place the ttf in data/Boldova.ttf.

Here are some examples (Left to right: Adele, Eels, Rage Against The Machine):

This project was a weekly project in the Data Science Bootcamp at Spiced Academy, April 2023.