Resume Parser with Natural Language Processing

Pornphat Sroison Assoc.Prof.Dr. Jonathan Hoyin Chan

School of Information Technology School of Information Technology
King Mongkut’s University of Technology Thonburi King Mongkut’s University of Technology Thonburi
Bangkok, Thailand Bangkok, Thailand
pornphat.phat@mail.kmutt.ac.th jonathan@sit.kmutt.ac.th

Abstract—Because of the advancement of the online II. OBJECTIVE

recruiting system. On the job application website, candidates
can easily upload their resume. Resulting in a huge number of 1) To use this technology that is based on natural
resumes being submitted. As a result, the human resource language processing to assist the human resource
department faces a challenge in recruiting new employees and department in screening resumes before conducting
reviewing a large number of resumes. Furthermore, interviews.
candidates who upload their resumes come in a variety of
2) To parsing and matching the similarities between a
formats, including writing style, fonts, font sizes, colors, and
etc. Human resource departments face a challenging problem candidate's resume and job description makes the hiring
in reading the entire resume that candidates upload and process easier and more efficient.
selecting the best candidate for the job position. So, for this 3) To help reduce human error and fatigue in screening
project, I propose to resume parser by using natural language resumes.
processing to assist the human resource department or
recruiter in extracting the detailed information of the resume III. SCOPE
that is needed to proceed with the applicant's process and also
Degree, field of study, and work experiences of
reduce errors in the work. This proposed system consists of
candidates are essential types of information for recruiting
three steps to parsing resume: 1) Receive resume files from
candidate 2) Convert resume file to the text format 3)
by the human resource department. They also want this
Extracting necessary information. The system will extract system to be able to rank or compare resumes to job
only relevant data that is necessary for the selection of the descriptions provided from them to evaluate if there are any
resume: name, consisting of first name and last name, position similarities. This will make it easy for them to work and
applied for, university, degree, skill, work experience, email make recruiting selections. As a result, where we have to
and phone number. In addition, the system can also display deal with a lot of data, converting a resume into formatted
the result percentage of similarity between resumes and job text or structured information to make it easier to review,
descriptions. To make it easier for recruiters to make analyze, extract relevant data, and understand is an essential
recruitment selection Keep your text and graphic files separate until after the text
has been formatted and styled. Do not use hard tabs, and
Keywords—Resume Parser, Extracting Information, limit use of hard returns to only one return at the end of a
Matching, Human Resource, Employer, Natural Language paragraph. Do not add any kind of pagination anywhere in
Processing the paper. Do not number text heads-the template will do
that for you.
I. INTRODUCTION
Nowadays, large companies and corporations have a Parse resume and match resume to job description
large number of candidates that are applying for jobs via are the two functions of this system. The first function is to
recruitment websites. Companies' human resource parse resumes. The user must upload a resume of the
departments or recruiters are responsible for screening candidate file in PDF or DOC format. This project supports
through a large number of resumes every day. This is not a only PDF and DOC format because they are the most
job for humans because screening numerous resumes and popular for creating resumes nowadays. The system will
selecting applicants for an interview takes a lot of time and read all text of the resume and extract only relevant data
can result in errors due to human fatigue. Resumes are that is necessary for the selection of the resume: name (first
unstructured data that differ from format of writing an name and last name), position applied for, university,
email, web pages content and others data with defined degree, work experience, skill, email address and phone
structure. Applicants' resumes generally include a variety of number. The second function is matching resumes to job
information, as well as colors, fonts, presenting order, and descriptions to evaluate how similar they are. The user can
literary styles in which they are written are different. upload a job description file and see the displayed result as
Resumes are also available in various file formats, including a percentage of similarity between resume of candidate and
'.txt,'.pdf,'.doc,'.docx,'.odt,'.rtf,' and etc. Those file types are job description. This system can reduce the HR
usually used by job candidates. As a result, an automated
department's time reading all text of a resume and reduce
intelligent system based on natural language processing is
errors in the work.
required to extract all of the information from unstructured
resumes and a variety of data sources. The method for parse IV. LITERAL REVIEW
resumes is converting all resumes to a similar structured
format and selecting only the information that is relevant to 1) Resume Analyzer Using Text Processing
screening, such as name, position, education, years of This literal review presents an effective Company
experience, work experience, certificates, email, phone Recommender System that uses text mining and machine
number, and etc. Following that, parsed resume data with learning algorithms to help recruiters select the best
structured format will be saved in a database for future use. candidate for a specific job. When candidates upload their
resumes, they are ranked according to the company's
requirements. The ranking can be used by the organization 1) Label is a label name that describes the type of
to select the best candidates. word.
2) Point Start and End is the number at the beginning
This article's methodology and model will be and ending position of desired word from all text in the
provided in four steps: collecting resumes and searching for resume after converting the PDF or DOC file into text
keywords stored in the information base on the resume text. format.
Then, ranking and Categorization of candidates based on a 3) Text is words in the content that is labeled
rating score. Furthermore, this system may extract new The entity name is a name about labeling or
keywords from resumes to expand the knowledge base tagging that we will classify the desired word that we have
further. specified position previously. In this project will use
specify the 2 entities' names: name and designation
2) Automated extraction of information from Polish
resume documents in the IT recruitment process VI. METHODOGY
This literal review analyzes and discusses This project implements Named Entity
automated information retrieval for the IT industry's Recognition, a part of Natural Language Processing that
recruitment process. In terms of low resource language analyzes large amounts of unstructured human languages.
dictionaries and complicated linguistic relationships in The initial step in extracting information and topic
Polish, the proposed approach implements a multi-module modeling is NER extraction. The system reads the whole
system. paragraph and highlights the text's key essential entity
This project uses the name of entity recognition, elements. Due to the resume text being an unstructured text
which is the most useful method for analyzing CVs. It's a into predefined categories, you can utilize Stanford NER or
semi-semantic analysis of the evaluated text that only Spacy for this project.
recognizes specific words. It's an essential phase in getting Regular expressions have been used in this
the text's information content ready for processing. project, as well as regular expressions in scripts. A regular
expression is a string of special characters that describes a
V. DATASET search pattern by matching a character pattern to the string
The data sets used in this project are divided into two being searched. Regular expressions consist of literal
datasets. First is a dataset of 200 resumes from GitHub symbols and special character combinations known as
consisting of names (first name and last name) and tokens, which indicate non-printable characters, symbols of
positions to apply for. Second is other datasets include a specific type, and the instructions for the regular
global university and skills. expression engine. It is a formal language theory and
theoretical computer science technique.
Table I. Number of datasets for each entity. A. PDF and DOC to text conversion
This project uses PyMuPDF library to convert
Entity Number of data PDF files to text format and python-docx library to convert
Doc, Docx file to text format.
Name 205
B. Named Entity Recognition (NER)
Designation 473 Extracting name (first name and last name) and
University 829 designation. This project uses the PKL or Pickle format for
Skills 1,249 the train dataset. Pickle is a Python module that serializes
objects so that they can be saved to a file and reloaded when
The train dataset for parsing consists of 2 parts. the program calls them. Then, uses Named Entity
The first part is content that specifies information about all Recognition (NER) for training model because this project
text of resume in the text format. The second part is an is a finding and classifying text of resume that is an
entity annotation that will be in the form: unstructured text into predefined categories by tagging
"annotation":{"label":["text"],"points":[{"start": dataset.
The number at the beginning of the word, “end”: The C. Regular Expression
number at the ending of the word, “text": "text of Extracting the name of university by using regular
content"}]}. The example of the train dataset for parsing expressions to search for keywords of university names
can be found on Fig.1 such as University, School, College, Institute, etc. After
that, searching for all the characters around those
keywords.
Extracting degree or educational background by
using regular expressions to search for keywords of
university names such as Bachelor of, Master of, Doctor of,
Degree, etc. After that, searching for all the characters
around those keywords.
Extracting skill. First step is cleaning data by
removing stop words that are a group of words that are
Fig 1: Example of the train dataset for parsing regularly used in a language but contain relatively little
valuable information, including punctuation on all text of
resume. Then, search for each token in the skills database
(.csv file). The final step is to create a bigram and trigram
from a string of tokens or a skill database, which are often
letters, syllables, or words, to identify a sequence of two or
three nearby parts.
Extracting experience. First step is cleaning data
Fig 4: The example results of cosine similarity score
by removing stop word and data preprocessing by word
with percentage that compares between resume of
tokenization. Then, parse regular expressions by using
candidate and job description.
chunk sequences of proper nouns ({<NNP>+}). The final
step is to search the word 'experience' in the chunk and then
VIII. LIMITATION
print out the text after 'experience' word in the line.
D. Regular Expressions in Scripts (Regex Scripts) Because of the data extraction limitations, it
Extracting phone number by using Regex Scripts includes some data that cannot be processed, such as the
to extract phone number: '[\+\ (]? [1-9] [0-9. \-\ (\)] {8,} [0- year of graduation and date of birth, which makes it
9]'. It works with standard phone numbers, including difficult to determine which class it is because the resume
country and area codes for most international numbers. mentioned many dates or years. In addition, there is not
Extracting email by using Regex Scripts to Extract enough dataset in this project, and the information
emails address: '[a-z0-9\.\-+_] +@[a-z0-9\.\-+_] +\. [a-z] +'. extracted does not cover all the details of the resume, such
It Works with all standard email addresses as long as the as experience. It can only retrieve a little amount of data
email uses standard English and @ characters. that is closely connected to the word "experience." As a
For the purpose of matching a resume to a job result, data retrieval problems are possible.
description that can be compared to see how similar they Resume parsing is also sensitive to ethical
are. The percentage of similarity will be displayed as the restrictions. Because of this system, the result will be a text
result. The way of comparison is importing a library from input only. As a result, this approach is only suitable for
scikit-learn (feature extraction) that can construct a count screening some positions. For example, a graphic designer
vector object to get a count of each word in the text and position or other design positions that require a visual
importing count vectorization. Then, using cosine preview of the work, an image as evidence of work, and
similarity, determine how similar two documents are. consideration of the resume's beauty and color may not be
appropriate for this system. This system's bias appears to
VII. RESULT be causing firms to lose employees.
The proposed system's results are shown in this
IX. CONCLUSION
part, which include extracting name, designation,
university, degree, skills, experience, email and phone Because the online recruiting system has
number using Named Entity Recognition to develop a progressed, a large number of resumes were submitted.
model and Regular Expression to extract the data. Another Consequently, hiring new employees and reviewing a large
feature of this system is that it compares the Resumes and number of resumes is a challenge for the human resource
job description of the applicant. The similarity of the department or employer. Therefore, this system has helped
outcomes is expressed as a percentage. Fig. 2, 3, and 4 employers by using an automated intelligent system based
show the entire system's results. on natural language processing. This system can convert
various formats of resumes to text format and can extract
some important information successfully. It is also possible
to compare the applicant's resume and the job description
to see the percentage of similarity as well. This system can
assist the human resource department or employer in
screening resumes before conducting interviews and
finding the best candidate for the job position.

Fig 2: The example results of parser resume that consist X. FURTHER DEVELOPMENT
of name, designation, university, degree, skills, This project intends to provide more datasets for
experience, email and phone number training in the future because the existing datasets are
insufficient for applications such as designation, university,
skill, etc. For future website development. This project will
A. Authors and Affiliations
apply the model to the website and add a function to view
the applicant's resume file or portfolio if the employer or
human resource department are interested. To support the
selection of resumes in all positions. After the user
confirms this candidate, the resume is saved in a NoSQL
database to be used as a future dataset, with the resumes
being ranked based on the percentage of similarity between
Fig 3: The example results of cosine similarity score the applicant's resume and the job description.
that compares between resume of candidate and job
description.
 To assist candidates, they can upload their
resumes to an online recruitment website to double-check
the information and compare the percentage of similarities
between their resumes and the job description to help them
decide whether to apply for a position.

REFERENCES
[1] What is resume parsing: Retrieved from,
https://www.smartrecruiters.com/resources/glossary/resume-
parsing/
[2] NLP Based Resume Parser using BERT in Python: Retrieved from,
https://www.pragnakalp.com/case-study/nlp-resume-parser-bert-
python/
[3] NLP based resume parser in Python (Beta): Retrieved from,
https://demos.pragnakalp.com/resume-parser/
[4] World University Ranking 2016: Retrieved from,
https://data.world/hhaveliw/world-university-ranking-
2016?fbclid=IwAR01WBDbntwc7K3NRkHpc1XCp8WcESQEV
MR2zXCXD8R31f-NTwJv1DZ7mWY
[5] Resume Parser: Retrieved
from, https://github.com/OmkarPathak/ResumeParser
[6] Resume and CV Summarization and Parsing with Spacy in Python:
Retrieved from, https://github.com/laxmimerit/Resume-and-CV-
Summarization-and-Parsing-with-Spacy-in-Python
[7] Automated-Resume-Screening-System Dataset: Retrieved
from, https://github.com/JAIJANYANI/Automated-Resume-
Screening-System
[8] How to extract email address, phone number and links from text:
Retrieved from, https://zapier.com/blog/extract-links-email-phone-
regex/
[9] Literature Reviews - Resume Analyzer Using Text Processing:
Retrieved from, https://jespublication.com/upload/2020-110557.pdf
[10] Literature Reviews - Automated extraction of information from
Polish resume documents in the IT recruitment process: Retrieved
from, https://www.sciencedirect.com/science/article/pii/S18770509
2101749X