Interviewing people is slow, costly, and inefficient.
- Slow because it is the slowest process in the whole Recruitment Process
- Costly: Consumes a lot of Recruiter’s time
- Inefficient: Mutual availability of the Job-seeker & recruiter simultaneously is an issue. Moreover, since the evaluation is done manually by a Recruiter, it can offer an unstandardized experience to Job-seekers, have bias & could be inaccurate.
Baoxu Shi, Shan Li, Jaewon Yang, Mustafa Emre Kazdagli & Qi He from LinkedIn have discussed making screening faster, inexpensive, and efficient in their research paper titled “Learning to Ask Screening Questions for Job Postings.” This research paper forms the basis of the following text.
Importance of this Research
Approximately 70% of phone screening find out that the applicant is missing basic qualifications such as years of experience in a specific domain, work permit, notice period, certifications, etc. If this screening is done in an automated way, it will help save precious recruiter hours and make the process faster & more accurate. This research paper lays the foundation to develop a two-stage deep learning model called Job2Questions which was widely successful and implemented by LinkedIn for Job postings. In A/B testing, the screening questions suggested by this DL model had a +53.10% acceptance rate, +22.17% job coverage, +190% recruiter applicant interaction, and +11 Net Promoter Score.
About this model: Key features of the questions generated by Job2Questions
- The questions could span a wide range of screening criteria covering experience related to a specific tool OR in a specific field, educational qualification, certifications, language competencies, or even availability.
- The DL model developed is generic to be applied across industries & roles. This model can also generate personalized questions based on the industry & role.
- Since the questions are objective, it helps remove Recruiter bias and makes screening via AI much easier & accurate.
- Low latency allows for real-time generation of screening questions.
- It saves time for Recruiters as applicants not qualifying the mandatory criteria can be rejected by AI. Moreover, since the decision on applicants’ candidature is taken in real-time (in case of rejection), it improves the Job-seeker experience.
- This model does not assume that the applicant’s social profile is updated, which is not usually the case. It also helps us avoid the fact that there could be a gap between what is mentioned in the applicant’s profile and what the recruiter expects.
Question template classification, Question parameter extraction, and question ranking are also explained in the research paper in detail.
Conclusion
In the words of the researchers,
In this work, we proposed a novel Screening Question Generation (SQG) task that automatically generates screening questions for job postings. We also developed a general candidate-generation-ranking SQG framework and presented LinkedIn’s in-production Job2Questions model. We provided design details of Job2Questions, including data preparation, deep transfer learning-based question template classification modeling, parameter extraction, and XGBoostbased question ranking. The extensive online and offline evaluations demonstrate the effectiveness of the Job2Questions model.
Source: Baoxu Shi, Shan Li, Jaewon Yang, Mustafa Emre Kazdagli & Qi He’s “Learning to Ask Screening Questions for Job Postings”
Discover more from Today Headline
Subscribe to get the latest posts to your email.