/Song_Sentiment_Ver2.0

Using the foundation and understanding of my first approach to song sentiment analysis using lyrics, created a more robust and better approach to this problem.

Primary LanguageJupyter Notebook

Song_Sentiment_Ver2.0

Using the foundation and understanding of my first approach to song sentiment analysis using lyrics, created a more robust and better approach to this problem.

Objective

Using song lyrics be able to tag songs based on the meaning of the lyrics without taking into consideration other variables such as genre, bps, or duration.

Resources:

  • Data Cleaning:

    • NLTK
    • Pandas
    • Numpy
    • JSON
    • REGEX (re)

  • Access to several APIs:

    • Spotify API and spotipy library in Python
    • Genius Lyrics API and lyricsgenius library in Python
    • Last.fm API
    • Wikipedia library in Python
    • AudioDB API and Beautiful Soup to extract the data

  • Modeling:

    • LDA Topic Modeling
    • Text Rank
    • Word2Vec

  • Testing Results:

    • Similarity Score
    • Confusion Matrix

Methods:

  • Pull songs from 24 popular playlists from the Spotify API.
  • Pull the lyrics of all the songs from the Genius Lyrics API
  • Pull the Genre of all the songs from the AudioDB API using Beautiful Soup
  • Use Wikipedia library to import genres and compare to LastFM tags
  • Combine all Data into a DataFrame
  • Use NLTK, REGEX and Spacy to clean up the lyrics
  • Use LDA Topic Modeling and Text Rank to create a taxonomy of the topics of each lyric
  • Use Word2Vec to vectorize the words in the lyrics
  • Using the vectorized words, calculate cosine similarity of each lyric to a topic in the taxonamy, highest similarity wins
  • Manually tag each lyric with one of the topics
  • Calculate accuracy of the word2vec model vs the manual labels