When several of Shirin Nilizadeh’s friends had lackluster outcomes in job searches, she knew something was up.
“I started asking why, as my friends are all qualified,’” says Dr. Nilizadeh, assistant professor of computer science and engineering. So she, along with doctoral student and project lead Anahita Samadi, decided to take a closer look. “What we discovered is that many times, resumes that contained certain keywords were being rewarded.”
Specifically, job recruiters use text-embedding to match words and sentences in resumes to the job description to obtain similarity scores and rank resumes accordingly. The study showed that job applicants can improve their position by at least 16 spots on average in a pool of 100 applicants by employing an algorithm that uses job-specific keywords. Few studies until now have shown that ranking algorithms that use text embeddings are vulnerable to adversarial attacks.
“We thought recruitment algorithms were the best example of such ranking algorithms,” Nilizadeh says. “Our goal was to identify the keywords from the job description that can improve the ranking of the resume.”
As expected, adding more keywords improves the ranking. The research also showed, however, that adding too many similar words or phrases might not improve the ranking of a resume.
“Therefore, we rank words based on their importance for a specific resume and then choose among the most important ones and add them to the resume,” she says. “In a way, it somewhat hacks the job application process.”
