How Machine Learning Drives TalentMarketplace

Published Dec. 11, 2019 by Scott Hirsch

This is the first of five posts about machine learning at TalentMarketplace. Links for each of the subsequent blogs are included in the headers below.

I've hired and been hired dozens of times. Each time is different and each time sucks.

We’re trying to make hiring suck less.

Over the course of 2019, we brought on a team of six data scientists to build upon TalentMarketplace (and its pool of 4000+ users) using the latest in machine learning & AI. Their results were too exciting not to share with you.

In this series of blogs, we will be going in-depth to see how these algorithms function. First, though, an overview of our four proprietary algorithms:

Candidate Resume Recommendation Engine


Moe, a Programmer with AQM and nearing the end of his second Bachelor’s degree, developed TalentMarketplace’s Candidate Resume Recommendation Engine. In Moe’s upcoming blog post, we explore how this recommendation engine enables employers to see lookalike candidates immediately upon viewing a candidate profile, saving them time. This model was made possible through the use of Word Mover's Distance, a data structure in a three dimensional space, which ranks similar resumes and shows the top three.

Candidate Summary and Suggested Skills Tool


Mukut, a Postdoctoral Fellow at TRIUMF experimenting on atomic parity nonconservation (APNC) measurement in francium atom and a Data Analyst at Applied Quantitative Methods (AQM), created the Candidate Summary and Suggested Skills Tool.

Mukut built a set of algorithms to train neural networks to understand the similarities between words (specifically skills) within a candidate profile. This skill mix is then used to predict the type of role the candidate is the best fit for. This provides employers with 1) automated summaries of candidates to employers to speed up their review times and 2) suggests key skills to candidates - skills the algorithm expects them to have, but the ones they did not mention - which are proven to help them get an interview.

Candidate Quality Estimator


Eric, a Graduate Research Assistant a PhD student specializing in physical chemistry and soft condensed matter physics, built the Candidate Quality Estimator. Using "an ensemble" of neural networks, this tool determines which candidate profiles are most likely to receive an interview request based on their profile metrics. The estimator assigns a score to each profile which is used to rank employers’ search results, saving employers time looking at potential candidates.

Candidate Interview Prioritization Tool


Nick, a PhD student and Simon Fraser University, worked on the Candidate Interview Prioritization Tool. This tool ranks candidates based on their likelihood of successfully completing an interview, based on the interviewers past decisions. This helps employers (and TMP internally) know where to spend their time. This tool is particularly unique as it could be repackaged separately, and used for high volume hiring in other industries.

All of these talented programmers have one thing in common: they all took on this intensive project and created a unique and intricate set of products that can help tailor an employer’s search to find the ideal individual for their firm.

Stay tuned through the months of January, February and March for a deeper dive into each machine learning tool and an understanding of how TalentMarketplace saves employers’ time in finding the perfect candidate.