Culture match

Team: Introversion

Track: Artificial Intelligence / Machine Learning

Video: Ultimate solution

Inspiration

I don't have previous experience of somehow fitting something to not clearly defined other thing. And actually I liked it, consider me inspired.

What it does

The first part of project just compares your cover letter with company descriptions on Wikipedia and returns the most relevant ones. You can use naive cosine similarity (it may well be the most reliable tool) or use a second model that predicts your corporate culture and compares with the corporate cultures of the company according to the Culture 500.

Challenges we ran into

I had to change the plan several times, not because there was a better option, but because there was not enough time. So frustrating! The main task was to find and process data, I spent too much time searching and now I have a lot of unused datasets.

Accomplishments that we're proud of

I found myself doing a good job of dealing with uncertainty, at least virtual.

What we learned

I had fun learning about how to get and process the multi-task output of a non-ANN model.

What's next for Culture match

• cluster employees according to their cover letters and select clusters in which there are already accepted ones
• use remaining data sources
• deploy as service
• evaluate in the wild

References

• Thumbnail photo: https://www.wsj.com/articles/five-office-designs-to-increase-productivity-1398387407
• English stopwords: https://countwordsfree.com/stopwords
• Culture 500 dataset: https://sloanreview.mit.edu/culture500/ scrapped with https://github.com/DarianNwankwo/culture500
• Tested on cover letters from: https://www.indeed.com/career-advice/cover-letter-samples
• Descriptions of companies scrapped from Wikipedia

Development tools: Unix as IDE, Geany, Python

Built With

• machine-learning
• natural-language-processing
• nltk
• numpy
• optuna
• python
• sklearn
• wikipedia
• xgboost