News Hub

LinkedIn’s new AI model automatically generates pre-interview screening questions

Written by Mon 4 May 2020

Social media platform has added an AI model to its job listings that significantly improves hiring efficiency

LinkedIn has developed an AI model for its jobs section that automatically generates screening questions for job listings.

In a paper uploaded to computer science pre-print server Arxiv, company researchers detailed the new Job2Questions application, which leverages deep transfer learning, a subcategory of machine learning, to generate appropriate screening questions for candidates.

Deep transfer learning is often used for applications where there is limited historical data to train a complex model. The researchers combined this technique with the ‘”Deep Averaging Network” model, a simple model favoured for its ability to quickly and accurately understand semantics.

The key idea behind the application is to improve hiring efficiency and automate time-consuming manual screening as far as possible. With Job2Questions recruiters can ask screening questions early on, enabling them to filter qualified candidates easily.

The AI model is currently active across LinkedIn’s 20 million job listings. When submitting a job listing, recruiters are asked if they want to add job questions (this has long been the case on LinkedIn), only now, the platform’s AI model suggests questions after digesting the job description, e.g. “How many years of work experience do you have using Java”.

To further boost functionality, LinkedIn developed another AI model to rank candidates based on the “favourableness” of their answers. LinkedIn found this had a marginal but significant impact on hiring efficiency, improving applicant “good fit rate” by 7.45 percent and reducing the bad fit rate by 1.67 percent.

Job applicants are also benefiting from the new features, the researchers said. When applying for jobs, LinkedIn now uses applicants’ screening answers to further calibrate job recommendations. When applying for jobs via email, LinkedIn users are 46 percent more likely to get a good fit rating if the job is suggested by the new and improved AI model.

It appears to be a win-win for both sides. According to LinkedIn, jobs with AI-generated screening questions lead to almost two times as many recruiter-applicant interactions and 2.4 times as many interactions with screening-qualified applicants.

LinkedIn’s AI team are not finished yet. The company’s researchers said in the future they plan to infer screening questions that are not explicitly mentioned in the job posting and investigate advanced question ranking methods to better model recruiter preferences.

Written by Mon 4 May 2020


deep learning linkedin
Send us a correction Send us a news tip