Resume parsing,ésumé also known as CV parsing, resume extraction, or CV extraction, allows for the automated storage and analysis of resume data. The resume is imported into parsing software and the information is extracted so that it can be sorted and searched.
Principle
Resume parsers analyze a resume, extract the desired information, and insert the information into a database with a unique entry for each candidate. Once the resume has been analyzed, a recruiter can search the database for keywords and phrases and get a list of relevant candidates. Many parsers support semantic search, which adds context to the search terms and tries to understand intent in order to make the results more reliable and comprehensive.
Machine learning
Machine learning is extremely important for resume parsing. Each block of information needs to be given a label and sorted into the correct category, whether that's education, work history, or contact information. Rule-based parsers use a predefined set of rules to parse the text. This method does not work for resumes because the parser needs to "understand the context in which words occur and the relationship between them." For example, if the word "Harvey" appears on a resume, it could be the name of an applicant, refer to the college Harvey Mudd, or reference the company Harvey & Company LLC. The abbreviation MD could mean "Medical Doctor" or "Maryland". A rule-based parser would require incredibly complex rules to account for all the ambiguity and would provide limited coverage.
This leads us to Machine Learning and specifically Natural Language Processing (NLP). NLP is a branch of Artificial Intelligence and it uses Machine Learning to understand content and context as well as make predictions. Many of the features of NLP are extremely important in resume parsing. Acronym normalization and tagging accounts for the different possible formats of acronyms and normalizes them. Lemmatization reduces words to their root using a language dictionary and Stemming removes “s”, “ing”, etc. Entity extraction uses regex expressions, dictionaries, statistical analysis and complex pattern-based extraction to identify people, places, companies, phone numbers, email addresses, important phrases and more.
Effectiveness
Resume parsers have achieved up to 87% accuracy, which refers to the accuracy of data entry and categorizing the data correctly. Human accuracy is typically not greater than 96%, so the resume parsers have achieved "near human accuracy."
One executive recruiting company tested three resume parsers and humans to compare the accuracy in data entry. They ran 1000 resumes through the resume parsing software and had humans manually parse and enter the data. The company brought in a third party to evaluate how the humans did compared to the software. They found that the results from the resume parsers were more comprehensive and had fewer mistakes. The humans did not enter all the information on the resumes and occasionally misspelled words or wrote incorrect numbers.
In a 2012 experiment, a resume for an ideal candidate was created based on the job description for a clinical scientist position. After going through the parser, one of the candidate's work experiences was completely lost due to the date being listed before the employer. The parser also didn't catch several educational degrees. The result was that the candidate received a relevance ranking of only 43%. If this had been a real candidate's resume, they wouldn't have moved on to the next step even though they were qualified for the position. It would be helpful if a similar study was conducted on current resume parsers to see if there have been any improvements over the past few years.