Content-based /Mood aware Recommendation system for Taylor Swift's Songs

Created a recommendation engine that recommends a Taylor Swift song based on the mood of the user.

• Gathered audio features such as (danceability and popularity) from the Spotify API.
• Scraped song lyrics from the web using Beautiful Soup.
• Made use of the NRC emotion lexicon to analyse the emotions expressed by the message of the song.
• Generated emotion scores for the lyrics of each song in the dataset.
• Used the audio features together with the emotion scores as the final set of features.
• Generated song recommendations using Cosine Similarity.
• Created a simple Streamlit Application to test out the engine.

Project Walkthrough :

Data Collection

metadata.ipynb Fetched Taylor Swift's music data using Spotipy a python library for the Spotify Web API.The audio features provided by the API include:

• acousticness
• danceability
• energy
• instrumentalness
• liveness
• loudness
• speechiness
• tempo
• valence
• popularity

taylor_lyrics.ipynb Fetched the song lyrics from the 6 most recent Taylor Swift albums namely RED,1989,REPUTATION,LOVER,FOLKLORE,EVERMORE, using beautiful soup.

Data preprocessing and Feature engineering

emotions.ipynb To Clean the lyrics I perfomed whitespace,punctuation,numbers and stopwords removal, lowercasing ,tokenisation and lemmatisation.Using nltk and pandas. Using The NRC emotion lexicon I created a dataframe showing the emotional inclination of each song against 8 emotions namely Anger Anticipation Disgust Fear Joy Sadness Surprise Trust.I then merged the audio features with the emotion scores into one dataframe.

Visualizing the emotional,lyrical and audio features present in her songs.

Taylor_analysis.ipynb

Making Recommendations

app.ipynb Generated item-item based recommendations using Cosine Similarity.Created a simple streamlit app to interact with the Recommendation Engine.