(https://arxiv.org/pdf/1708.07903v1.pdf) This paper guided this approach to predicting ethnicity based on a resume. The goal is to use the principle of homiphily (people associate with others of similar ethnicity) to predict ethnicity.
- dataProcessing.py - Given a directory of json resumes, creates two dictionaries containing the information we need. (name_to_related - goes from names to urls of related people on linkedin) (url_to_name - goes from urls to names)
- randomDataSample.py - Given the outputs from dataProcessing.py, creates a random sample of people and their connections from a seed number. It only takes people who have connections (~1/20). Essentially is a depth 2 search and will create pairs of people. Creates an adjaceny list from a the randomly sampled people.
- decisionRule.py - Given an adjaceny list, creates a graph and labels all people via: (US Census, Character Set, Homiphily)
- Rerun the process on the newer version of the data. See if there is a higher rate of connections.
- Train a more traditional classifier based on the resumes, by hand selecting features that do not relate to employment.