Building a Character Recognition Pipeline with a frontend interface

[fyang3]

The character class in the gender_analysis toolkit provides the functionality to automatically generate a character list with each character’s name, nicknames, and pronouns based on a particular document input and intake user feedback for a manually disambiguated list. The pipeline utilizes a human-AI collaboration approach that includes NLTK’s Named Entity Recognition (NER) and Neuralcoref’s Coreference Resolution model as well as a manual disambiguation interface. For the gender analysis web interface, we’d like to build a frontend that achieves the core functionality of the pipeline:

MVP:

• A user selects a document through leveraging our document model
• The backend pipeline automatically output a list of character names with their associated nicknames and pronoun probabilities based on THIS_NOTEBOOK
• A frontend disambiguation interface that enables the user to validate and correct the pipeline outputs through a dropdown list design (or similar)

Nice-to-have:

• Output a resolved text with the results from the character identification-disambiguation pipeline
• Take the resolved text for further analysis similar to proximity analysis and frequency analysis

[kenalba]

Working on this here:

https://github.com/dhmit/gender_analysis_web/tree/character_alias_model