Automating the Job Hunt with Transfer Learning Part 1

Today, we are closer than ever to automating everything to give us more time to learn even more cool things. We need a one click job seeker.

“Problems are nothing but wake-up calls for creativity” — Gerhard Gschwandtner

Introduction

I have started a mid-career transition from high school teacher to data scientist and must admit, the longer the search goes on, the more disheartening it feels. Instead of getting down on myself and giving up, I saw an opportunity to grow. You didn’t come to read about job search struggles, but you came here to read about how I used data science to reduce those struggles. So, in this series of articles you will get a quasi-high-level explanation and walk-through of a portfolio project that aims to use text generation to write cover letters and maybe even your resume.

First, this article is an introduction and rationale behind the creation of this model. Maybe it’s blowing off some steam too.

The second part of the series involves extracting the desired skills from the job description. We will use POS tagging and chunking to regex the skills and non-skill phrases from the dataset. In this part 2.

The third part will use the BERT (Bidirectional Encoder Representations from Transformers) model word embeddings in conjunction with grammatical patterns in English sentences to extract and classify skills. Link to come;

Lastly, we utilize these extracted skills to develop a cover letter using the GPT-2, maybe newly released 3, (Generative Pre-Trained Transformer) model. The key to this last part is altering the embeddings to ensure the desired skills are generated within the text. Link to come;

One of the first tenets one learns in data science is optimization; every part of data science boils down to efficiency. With my newfound data science knowledge and my innate disposition of efficiency, in other words efficient laziness, I wanted to kill two birds with one stone.

1. The first bird dealt with filling my portfolio with meaningful, original and fun projects. I am reading about how a portfolio is better than a resume. This portfolio should have projects that show my best qualities: creativity, problem solving and ambition. These can’t be gleaned from a resume but if someone reading your cover letter or resume learns that:[A. it was automatically generated; B. it has contextual similarity to the job description; C. this is a product of my data science skills] then it shows an ability to deliver an optimal solution.
2. The second bird dealt with the time and planning/effort involved with the job hunt. How many times have you heard “Job hunting is like a full-time job?” Imagine writing a different cover letter for every job you applied to; you’re supposed to! That is a lot of cover letter writing and research to do about the company. Also, being a data science generalist, you could have a resume that focuses on data analysis, data engineering, NLP, etc. You should also have a resume outlining those specific skills. There is a great deal of planning for an effective job hunt.

Get the best ROI (return of investment) by reducing the “cost” of applying.

For awhile, I made individual cover letters and continuously updated and tweaked many different resumes with no success of landing a full-time job. I managed to get an internship that was great and a few small contract positions for some freelance work but, never that satisfying full-time role. Perhaps my resume was not getting past the ATS (Applicant Tracking System), I had too little experience or because it must be remote because I am stuck on a small rock in the Caribbean. (Although it’s not a bad place to be!). Maybe, well probably, I had a typical cover letter and a resume that didn’t quantify or highlight particular skills. Whatever it is, I want to fix it.

What if, this is probably getting into a grey area, you had the companies coming to you? My rationale behind this is k-fold ;). Not only do I optimize this problem, but I am turning the tables to a certain extent. Ultimately, the idea is to minimize time spent on those who will not reach out to proceed with an interview process. When you get a “hit” (a reply from a company after applying), you now have an opportunity to research the company in detail to make a decision on agreeing to an interview process. Using the law of large numbers, I suppose you will get better results. When you write a resume and cover letter tailored to a specific company, that time you have spent has a cost. The cost is different for different people but for me it takes time away from doing two things I enjoy the most: playing with my kids and learning/doing data science. That’s what it all boils down to.

This project is an experiment in job hunting and utilizes an artificially intelligent model and a few statistical concepts in an attempt to make job hunting a piece of cake. Check out the next part in the series.

If you are interested in following this series of articles and many of the other projects I am currently working on please subscribe to my newsletter here. You will get project ideas, tutorials and stay up to date on the most current news in data science.