Does Your Resume Parser Stack Up? How to Evaluate Next-Generation Systems

By Dave Zielinski May 10, 2016

The arrival of resume parsing technology heralded a new day for recruiters.

Automation saved talent hunters precious time and spared them headaches by extracting relevant data from candidate resumes and placing it in searchable databases.

But today those original parsing methods, which relied heavily on counting resume keywords, are growing dated. If the parsing software in your applicant tracking system (ATS) doesn’t go beyond mere keyword matching to reading resumes more like a human would, you're missing out on important new capabilities, recruiting experts say.

The new generation of parsers uses what's known as contextualized search to evaluate keywords within the context of entire resumes. This allows the software to more readily identify factors like how old skill sets are or how the responsibilities included under a job title like “associate” or “vice president” might vary from company to company.

Such natural language processing can help distinguish between a candidate who took classes in WordPress software a few years ago, for example, and someone who's actually used the tool on the job for the past three years. 

State-of-the-art parsers also can be configured for atypical recruiting needs, will accommodate most languages and offer software-as-a-service (SAAS) delivery models.

Data Normalization Improves

Any parser needs to execute two tasks well. It should rapidly and accurately extract designated data from resumes beyond contact information or work histories to include things like desired location, candidate summaries or even visa status. The software should also allow for easy searching by analyzing extracted information that’s placed in resume databases.

“The breadth of data fields a parser can fill beyond name, rank and serial number is important,” said Matt Sigelman, CEO of resume parsing company Burning Glass Technologies in Boston. “The more information that is accurately segmented and structured, the easier it is for people to pinpoint the information they’re looking for, whether it be candidates who’ve worked at certain companies or have specific certifications.”

Contextualized resume parsing also reduces the odds of misclassifying words on resumes that have multiple meanings, which can throw a wrench in database searches, according to Sigelman.

“You may be looking for someone who is experienced in using Harvard Graphics versus someone who graduated from Harvard, or someone who’s programmed in Java rather than worked at Java Joe’s coffee shop,” Sigelman explained. “You want to be able to exclude keywords used in a context that’s irrelevant to your search.”

Time-based tagging of resume data by parsers also allows recruiters to search for candidates who’ve had specific words in their recent job titles—such as “Ruby on Rails programmer”—not just their titles from years ago. 

Accuracy remains a key measure of parser effectiveness.

“What you’re looking for is close to 90 percent parsing accuracy overall on traditionally formatted resumes as well as an ability for the technology to normalize data very effectively,” said Elaine Orler, founder and CEO of Talent Function, a recruiting technology consulting firm in San Diego. Orler is also a former member of the Society for Human Resource Management (SHRM) Technology and HR Management Special Expertise Panel.

Systems that don’t normalize data well might take a misspelled company name (like “MacDonald’s”) and fail to correct it before placing it in a data field, Orler said. That means recruiters conducting a search for the name wouldn’t find it. 

Parsers need to go beyond accurately filling in ATS forms or website applications with resume data, said Gerard Mulder, chief commercial officer for Textkernel, a parsing company based in Amsterdam that’s owned by the job board CareerBuilder.

“A parser needs to classify data, enrich it with knowledge from other sources, normalize data so it can be used for analysis and allow for better searching,” Mulder said.

Comparing Parsing Providers

Any evaluation of parsing software in an ATS should include real-world tests, experts say. Robert Ruff, CEO of parsing company Sovren in Houston, said vendors “tend to blow a lot of smoke” and the only way to gauge performance is through your own trials.
“Test 15 of your own resumes against different parsers and see what you find,” Ruff said.

While resume parsing is simple in theory, Ruff said what separates the best parsers from others is a commitment to continuous improvement.

“There are probably five times as many resume parsing companies that have gone out of business than exist today,” Ruff said.  “Most went out of business because they stopped doing the hard work of making their product better every single day. You can spend the rest of your life trying to create the perfect parser.”

Does the Parser Flex?

Another key measure of a parser is its ability to adapt to atypical resumes or recruiting needs. Resumes can feature specialized formats, tables or other information that diverges from the norm. One client of Edinburgh, Scotland-based parsing company Daxtra wanted to be able to distinguish candidates’ military history from standard work history, said Steve Finch, CEO of Daxtra, and the company accommodated the request.

Sovren’s system also can be configured to meet custom requests. “Let’s say you’re recruiting in Bolivia and there are some unique names used for educational degrees in that country that aren’t used elsewhere,” Ruff said. “We can program our parser to search for those different degree names.”

Foreign Languages and Social Profiles

Organizations doing business globally also need systems that can parse in multiple languages. Top parsing providers also understand that the same language can be used differently depending on geography. A parser developed for the use of German language resumes in Germany might differ from one built for German use in Switzerland, for example.

When parsing provider HireAbility received a request from a job board in Romania for its services, the company quickly asked for 100 resumes in that language to train its software, said Steve Kenda, CEO of Hireability in Londonderry, N.H.

“We also trained the parser for clients in Bulgaria, Slovakia and Greece who were receiving Romanian resumes as well,” said Kenda.

As more companies allow candidates to apply for jobs with social media profiles, parsing vendors have also had to adapt to accommodate the new formats.

“Parsers today should be able to extract information about candidates wherever their profiles exist on the Internet, be it on LinkedIn or on an industry-specific site like GitHub,” said Finch.

Automation Dangers?

Some wonder whether these advances in parsing technology will have unintended consequences.  When there is less reliance on humans to interpret resume data, for example, might it increase the chances that top candidates will be overlooked? Parsing vendors, quite naturally, don’t share those concerns.

“We believe using the technology takes away that risk rather than increases it,” said Mulder of Textkernel. “Combining contextualized parsing and semantic search-and-match technology increases the chances that those candidates will be found by recruiters because the software can suggest what candidates mean rather than just what they’ve written.”

Dave Zielinski is a freelance journalist specializing in HR technology, talent acquisition and leadership development issues.



Hire the best HR talent or advance your own career.

Discover what’s trending in HR

Search and download FREE white papers from industry experts.

Search and download FREE white papers from industry experts.



HR Daily Newsletter

News, trends and analysis, as well as breaking news alerts, to help HR professionals do their jobs better each business day.