Book narratives

• follow the sentiments in books over time (much like twitter and other word data)
• from data emerges consistent story arcs
• see slides

Machine learning in book data (words)

• history of understanding ratio of male:female centered books
• understanding gender bias requires intensive categorization / reading
• use information from 200 books and features about those books to determine important features in predicting whether the book is male or female centered
• selecting important features (from a long list) is the same as selecting important predictors in marketing (ie who will buy) or in recidivism (who will reoffend)
  • see slides
• test model on another set of books and examine success
• predict male or female centered for all remaining books

Grouping Sherlock Holmes books (see the app)

• without you providing a list of features, ML groups books based on word frequency
• unsupervised ML (compared to supervised ML, where characteristics were compared to known outcomes)

Where are you from? (see the app)

• pretty accurate
• based on multiple questions each of which provides some probability of assigning you to a location
• questions may differ for different people as particular sets of questions indicate particular locations
• each question has some info, no one answer is diagnostic

Lyrics (see interactive site)

• some lyrics are timeless (Love)
• repetition is increasing
• number of words is decreasing
• vulgarity is increasing
• word use reflects society
• ML can be used to develop a model to predict genre from lyrics

How we find music has changed

• what music you listen to is the result of targeting (eg Spotify suggestions) rather than general play