Quick summary:
- What I did:
- built deep neural architecture to train embeddings on a dataset of job ads and skills
- Things I learnt:
- how to use a computing cluster to train deep neural models
- When was this?
- September 2019 through December 2019
I had the opportunity to work with Prof. James Evans at the University of Chicago on an analysis of tens of GB of job advertisement data.
The idea was to explore the relationships between skills based on how they were talked about in job ads. Some skills are general (‘metalworking’), while others are specific (‘TIG welding’). Some skills are precursors to others (like ‘event organizing’ and ‘event management’). Some occur frequently with certain other skills (‘software development’ and ‘Java’ or ‘marketing’ and ‘public relations’) while other combinations are rarer. Can we infer these types of semantic relationships between skills from data? Can we predict the births of new skills, or the deaths of old ones?
Data analysis has the power to capture these multifaceted relationships in elegant ways based on simple first principles.