<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=1882897988680532&amp;ev=PageView&amp;noscript=1">

Jobful Blog - Develop Your Career!

A new & powerful tool in your recruiting arsenal: predicting candidate response likelihood

Jun 23, 2020 9:44:59 AM / by Maria Ivanciu

This week we are happy to announce that we are incorporating a brand-new algorithm into our gamification platform to calculate the user’s response likelihood. On this occasion, we would like to give you a glimpse into our cutting-edge technology and, as usual, make the recruitment process easier for you.

 

technology

 

Algorithm Description

Starting from the idea of predicting whether a user will respond to a job invite from a certain company, our algorithm implements machine learning techniques based on a probabilistic approach. As input, the AI takes into account users’ behaviour on the platform and then it builds a model from users’ profiles.

 

How does it work?

Essentially, the model assigns scores based on certain criteria (detailed below). The response likelihood score is estimated and extracted from the data while the model is trained to learn what characteristics are negative and which ones are positive.

For instance, if the user used the platform in the past 48 hours, he / she  receives a small positive score. This can be further increased if the user completes a Jobful Academy. On the other hand, if the user didn’t complete their skills, the score drops.

Finally, the output is the model’s response to the question: “If company Y sends a job invite to user X, what is the probability that user X will respond?”. If the probability is above a certain threshold, company Y will capture a “high” response in user X’s application.

 

Data Sources

As any other machine learning model, our response likelihood algorithm relies on data generated by our platform, carefully processed such that it would not create bias towards certain users, e.g. age or gender. In other words, we only use non-confidential and anonymised data such that the model picks up only relevant features.

 

By no means an exhaustive list, the data streams from:

  • User’s engagement in the past month on the platform: logins, CV updates, challenges, academies, job searches, visited jobs
  • Job search: applications, invitations, unanswered invitations
  • Performance: CV percentage, points, chest, abilities, badges, tests
  • Professional development: skills, work experience, education, volunteer work, projects, accomplishments, languages
  • Information about the relation with the company: company visited jobs, invitations, unanswered invitations from the company

Each of these items listed above contributes with some weight towards the final score.

 

artificial-intel

!!Disclaimer

The aim of the model is to indicate whether a certain user may respond to a job invitation from a specific company. It does NOT represent a decision on the success of the application and it does not interfere with the recruiter’s decision. The model’s functionality is intended as a new tool within the recruitment process and not as the central criteria for candidates’ success.

 

Taking into account that the model is based on a decision tree approach, the prediction may be wrong at times.

However, the beauty of this model is that every week it becomes better by revising its errors and by re-learning from these mistakes. This improves the accuracy and precision of the predictions. However, keep in mind that nothing can predict the future with maximum accuracy.

Play your way to a killer job and enjoy your journey with Jobful! 

Tags: Product Updates

Maria Ivanciu

Written by Maria Ivanciu