You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We could build a method that takes in a Document, detects whether or not it's an epistolary novel, and then breaks the document up into a dictionary of letters (or a list of Letter objects?). We'll want to programmatically detect the writer of each letter and include that in the metadata.
Ideally, we can programmatically determine metadata for each letter - writer, date, recipient, and so on. That's going to be tricky, but maybe possible. If we combine this functionality with our hypothetical named entity recognition module (to get a character list) and a ML-based gender guesser for each character, we can do some classy stuff.
Activity
fyang3 commentedon Oct 27, 2020
https://www.mygreatlearning.com/blog/named-entity-recognition/. A pretty good overview for Named entity recognition.
Microsoft Azure also has NLP modules on this: https://docs.microsoft.com/en-us/azure/machine-learning/studio-module-reference/named-entity-recognition
For detecting gender by names we could use NLTK or Scik-learn and build our own classifier (so we need to decide what features we'd like: https://www.geeksforgeeks.org/python-gender-identification-by-name-using-nltk/
This is an example of building up a classifier: https://gist.github.com/vinovator/6e5bf1e1bc61687a1e809780c30d6bf6