Is Boolean Search Dead for Recruiters?
Dr. Yihua Liao is the Chief Data Officer and Co-founder of Brilent. He oversees the research and development of Brilent’s big data and artificial intelligence (AI) based recruiting technology. Before joining Brilent, he was a data science manager at Facebook, leading a team of data scientists in multiple product areas including payments, games, commerce, and site integrity. Prior to Facebook, he developed identity and payment fraud detection systems using machine learning techniques at ID Analytics and Microsoft. He holds a Ph.D. in computer science from the University of California at Davis.
Most recruiters are familiar with Boolean searches in the context of looking for candidates, and I’m going to talk about Boolean search in that context in just a moment. But before I talk about recruiting, let me tell you a story about Boolean search that happened to me in a different industry. Humor me for a moment and you’ll see how this anecdote is relevant to recruiters!
A tale of Boolean search
It was the fall of 2006. I had just joined Microsoft as a Decision Scientist (back then the title “data scientist” had not been invented yet!). My job was to mitigate the payment fraud that was surging on their online services platform that supported Xbox Live, Office Live, and many other products. I worked with a fraud investigation analyst, let’s call him Stirling, side by side. Before I joined the team, Stirling had been asking software engineers to run Boolean searches and pull payment transactions for him to review. A typical Boolean search string would look like this:
All Xbox Live transactions with amount greater than $50 AND (account was created within last 7 days OR IP was from outside of US)
Then Stirling would manually review every transaction the Boolean search returned and try to decide whether they were fraudulent or not. Boolean search worked well initially when the transaction volume was low. But as the business grew, Stirling quickly became overwhelmed.
The limitations of Boolean search were pretty clear to us:
It was impossible to come up with the perfect Boolean string, especially when the fraud patterns got more sophisticated. It was either too restrictive (too few transactions were returned) or too lenient (too many transactions met the Boolean conditions).
- There was no ordering in the transactions returned. Stirling had to go through all the transactions one by one, which was very tedious and time-consuming.
With Stirling’s help, I quickly implemented an AI and machine learning-based fraud detection system that incorporated much of his domain knowledge. The system generated a fraud score for every transaction, representing how likely the transaction was to be fraudulent. Then Stirling would only review transactions with the highest fraud scores. As a result, he reviewed less transactions every day, but he was able to catch more fraud. Together the team significantly improved productivity and lowered the fraud rate. Over time, the accuracy of the system got better and better, and we never looked back on Boolean search.
Boolean search for recruiting
Fast forward to 2017. I am now working at a recruiting software company. What I’ve learned is that Boolean search is still widely used in the recruiting industry. Talent seekers heavily rely on Boolean search to find candidates in different systems such as LinkedIn, applicant tracking systems (ATS), CRMs, and job boards (with some minor variations). They are having similar frustrations to what Stirling experienced 10 years ago, because of the inherent limitations of Boolean search as well as the unique challenges in recruiting. Let’s take a look at a few of those limitations.
Lack of standardization and knowledge base of basic entities such as skills and job titles. By its very nature, Boolean search tries to find in a candidate’s resume or profile the exact terms specified in the query. For example, searching the skill “Transact-SQL,” and you will find candidates with the exact skill. A smart search should include the synonym “T-SQL” and other relevant skills such as “SQL” and “MySQL.” Similarly, search for candidates with the title “Web Developer,” and you will miss similar titles “Front-end Engineer” or “Front-end Developer.” Most of the systems today either don’t have the built-in standardization and knowledge base at all, or require a fair amount of manual tweaking.
Lack of Boolean dimensions that you need to describe the right candidates. Have you tried to search candidates who have worked at their current job for more than 4 years or have relevant experience in a particular field (e.g., digital marketing) for more than 5 years? Questions like these are critical to hiring managers and recruiters. Yet you can’t query in those dimensions on even the most advanced Boolean search systems today.
No particular ordering and ranking of candidates in the Boolean search results. A simple search, “Software Engineer” in the “San Francisco Bay Area” that I ran on LinkedIn this morning, returned over 152k candidates. Even after I added additional filters on skills and seniority level, I still ended up with 2,874 candidates. Because there was no particular ordering of the candidates, I would have to go through all these candidates in a brute force fashion in order to find the right ones.
Due to these reasons, many recruiters are looking beyond Boolean search and ready to move on to new tools that can make their jobs easier.
The age of artificial intelligence (AI)
By now most of us are familiar with intelligent systems that recommend books, movies, music, news, and other products to consumers. In the enterprise world, artificial intelligence has also significantly increased the productivity of workers in sales, marketing, financial services, and many other industries. We believe artificial intelligence can help people in the recruiting industry as well.
An intelligent candidate recommendation system should be able to handle many tasks intelligently as a human recruiter would, in a much more efficient and scalable way. Specifically, it should:
Understand job requirements from a job description, including job title, seniority, location, desired experience, and required or preferred skills.
Evaluate candidates holistically, the same way a recruiter does, by considering and weighing all of the factors that matter, such as education, experience, skills, career trajectory, and so on.
Understand the relationships of different entities, such as skills, job titles, companies, and schools, and other common knowledge that a recruiter has in their head.
- Stack rank candidates for job fit so that recruiters can identify their top candidates immediately.
At Brilent, we have built just such a recommendation engine for recruiters. Other tech companies are also entering the recruiting space with their AI-powered recruiting tools, including IBM's Watson for Recruiting, Google's Cloud Jobs API, and Mya, the recruiting chatbot.
AI for recruiting is already here! Maybe I was a bit hyperbolic in the title of this post. Boolean search will probably still be around for a while. But it shouldn’t be your best friend anymore. AI is, or will be soon. Just as it helped my coworker Stirling 11 years ago, it has the potential to save time and make recruiters much more productive—and that’s something we should all be excited about.
Want to keep up on the latest news and trends in talent acquisition? Sign up for The Modern Recruiter newsletter!
Please note: This guest post was written by our partners at Brilent and does not necessarily reflect the opinions of Greenhouse.