Hireability: Information Extraction from Resume

Student Id : 2050196
Student Name : Prasuna Pokharel
Abstract

Agencies and various high-level firms must deal with a large number of new jobs seeking
people with various resumes. However, managing large amounts of text data and selecting
the best-fit candidate is more difficult and time-consuming. This paper provides an overview
of an ongoing Information Extraction System project that helps recruiters in identifying the
best candidate by extracting relevant information from the resume.

This project presents a system that uses Natural Language Processing (NLP) techniques to
extract minute data from a resume, such as education, experience, skills, and experience.
The recruiting process is made easier and more efficient by parsing the resume. The
proposed system is made up of three modules: an administration management system, File
upload and parser system, and an information extraction system. The administrator will
upload the applicant's resume into the system, and the relevant information will be
extracted in a structured format. Using the parsed information from the Resume, HR can
select the best candidate for the job based on the company's needs.
 Contents

1. Introduction..................................................................................................................................

1.1. Problem Statement...............................................................................................................

1.2. Implementation of module....................................................................................................

1.2.1. Reason behind implementing NLP................................................................................

1.2.2. Work Flow of NLP..........................................................................................................

1.3. Aims......................................................................................................................................

1.4. Objectives.............................................................................................................................

1.5. Artefacts................................................................................................................................

1.6. Scope and Limitations of the project....................................................................................

1.7. Structure of the report..........................................................................................................

1.8. Academic Questions.............................................................................................................

2. Literature review..........................................................................................................................

2.1. End-to-End Resume Parsing and Finding Candidates for a Job Description using..........

BRET.........................................................................................................................................

2.2. Information Extraction from Free-Form CV Documents in Multiple Languages................

2.3. Resume Information Extraction with a Novel Text Block Segmentation Algorithm...........

2.4. Information Extraction from Resume Documents in PDF Formats....................................

2.5. Study of Information Extraction in Resume........................................................................

2.6. Automatic Extraction of Usable Information from Unstructured Resumes to Aid..............

Search........................................................................................................................................

2.7. Analysis and Findings.........................................................................................................

2.8. Conclusion of literature review...........................................................................................

3. Project Methodology.................................................................................................................

3.1 Scrum Methodology.............................................................................................................

3.1.1. Reason behind Selection.............................................................................................

3.2. Initial Gantt Chart................................................................................................................

3.3 Additional info......................................................................................................................

3.3.1 Resources.....................................................................................................................

3.3.2 Client.............................................................................................................................

4. Technology and Tools...............................................................................................................

5. Artefact Design..........................................................................................................................

5.1. System Requirement Specification....................................................................................

5.2. Functional Decomposition Diagram...................................................................................

5.3. Legend................................................................................................................................

5.4. Admin Management System (AMS)...................................................................................

6. Diagrams...................................................................................................................................

6.1. Activity Diagram..................................................................................................................

6.2. Use Case Diagram.............................................................................................................

6.3. Sequence Diagram.............................................................................................................

6.4. Wireframe Designs.............................................................................................................

6.4.1. Home Page..................................................................................................................

6.4.2. Login Page...................................................................................................................

6.4.3. Parsing Page................................................................................................................

6.4.4. Information Extraction Page.........................................................................................

6.4.5. FAQ Page.....................................................................................................................

 6.4.6. Terms and Condition Page..........................................................................................

6.5. Design Process...................................................................................................................

6.6. Development.......................................................................................................................

6.7. Product Backlog..................................................................................................................

6.8. Accessibility........................................................................................................................

6.9. System Work Flow..............................................................................................................

6.10. Testing and Evaluation.....................................................................................................

6.10.1. Accessibility Testing...................................................................................................

6.10.2. Black Box Testing......................................................................................................

6.11. Module Implementation....................................................................................................

6.11.1 Important Necessary Libraries and Module................................................................

6.11.2. Data Collection...........................................................................................................

6.11.3. Data Pre-Processing..................................................................................................

6.11.4. Code and its explanation............................................................................................

6.11.5. Output.........................................................................................................................

7. Conclusion.................................................................................................................................

8. Critical Evaluation......................................................................................................................

8.1. Final Report........................................................................................................................

8.2. Finding and process...........................................................................................................

8.3. System................................................................................................................................

8.4. Future Work........................................................................................................................

8.5. Self-reflection......................................................................................................................

9. Evidence of Project Management.............................................................................................

 9.1. Gantt Chart.........................................................................................................................

9.2. Project Management tool....................................................................................................

10. References..............................................................................................................................
Figure 1: Single Column Resume ........................................................................................ 2
Figure 2: Double Column Resume ....................................................................................... 2
Figure 3: Free Format Resume (Image)............................................................................... 3
Figure 4: Generic Workflow of NLP ...................................................................................... 5
Figure 5: A system diagram showing data transmission and task completion. .................. 11
Figure 6: The Information Extraction System from Free-form CVs: A High-Level Overview
........................................................................................................................................... 12
Figure 7: Workflow of Information Extraction from Resume ............................................... 13
Figure 8: Workflow of the System ...................................................................................... 15
Figure 9: Skillate Software ................................................................................................. 17
Figure 10: DaXtra Software................................................................................................ 18
Figure 11: Sprint Backlog ................................................................................................... 20
Figure 12: Initial Gantt Chart .............................................................................................. 20
Figure 13: Functional Decomposition Diagram .................................................................. 23
Figure 14: Activity Diagram of admin management system ............................................... 25
Figure 15 activity diagram of parsing system ..................................................................... 26
Figure 16: Use Case Diagram ............................................................................................ 27
Figure 17: Admin Management System ............................................................................. 28
Figure 18: File Upload System ........................................................................................... 28
Figure 19: Design Part ....................................................................................................... 35
Figure 20: Invert Color ....................................................................................................... 37
Figure 21: Increasing and Decreasing Text size ................................................................ 38
Figure 22: System Work Flow ............................................................................................ 38
Figure 23: Light House Testing .......................................................................................... 39
Figure 24: Color Contrast Test ........................................................................................... 40
Figure 25: Screenshot of Test 1 ......................................................................................... 41
Figure 26: Screenshot of test 2 .......................................................................................... 42
Figure 27: Screenshot of test 3 .......................................................................................... 43
Figure 28: Screenshot of test 3 .......................................................................................... 44
Figure 29: Screenshot of test 4 .......................................................................................... 46
Figure 30: Tokenization...................................................................................................... 50
Figure 31: Stop words ........................................................................................................ 51
Figure 32: Lemmatization................................................................................................... 51
Figure 33: Lower Casing .................................................................................................... 52
Figure 34: Extract the text from pdf .................................................................................... 53
Figure 35: Extract Text from Doc files ................................................................................ 53
Figure 36: Detect the file extension .................................................................................... 54
Figure 37: Extract the entities ............................................................................................ 54
Figure 38: Extract the email ............................................................................................... 55
Figure 39: Extract Name from Resume .............................................................................. 56
Figure 40: Extract Mobile Number ..................................................................................... 57
Figure 41: Extract Skills from the resume .......................................................................... 57
Figure 42: Extract Technical Skills ..................................................................................... 58
Figure 43: Extract Education and Year .............................................................................. 59
Figure 44: Extracting address ............................................................................................ 59
Figure 45: Extract Experience ............................................................................................ 60
Figure 46: Output ............................................................................................................... 60
Figure 47: Final Gantt Chart............................................................................................... 81
Figure 48: Project Management ......................................................................................... 81
1. Introduction
Daily, corporate firms and recruiting agencies have to process a large number of resumes.
Working with a large volume of text data is usually time consuming and stressful. Data
gathered from different resumes can be in a various form, including .pdf, .docx, single
column resumes, double-column resumes, free formats, and so on. And these formats
might not be suitable for the particular application. So, questions may arise in our mind
that, what is resume parsing? The process of converting the unstructured form (.pdf/ .docx
/ .jpeg etc.) of resume data into a structured format is known as resume parsing.

Subsequently, converting a resume into prepared text or structured information makes

studying, analyzing, and comprehending easier. As a result, many organizations and
institutions depend on Information Extraction, where unstructured data and vital
information are extracted and converted to make information more readable and
organized data forms. The completion of this task takes a long time for humans. So, it is
necessary to develop an automated intelligent system that can extract all relevant
information to determine whether an applicant is suitable for a particular job profile
(Kurama, 2021).

The foundation of this project is a resume automation system. Concerning the project,
there will be an admin panel at first, into which the administrator must initially log in.
Following that, there will be a section for uploading gathered CVs, which the admin will
manage. After that, Regex, NLTK, and Spacy's phrase matcher will extract necessary
information such as Name, Address, Email, phone number, Nationality, Skills [Hard Skills,
Soft Skills], Education, Experience, Experience Year, Languages etc. in. json format.
Finally, the extracted information or dump JSON file is saved in the database by admin if
necessary.
Below is the wireframe of Unstructured Data (Single-Column, Double-Column, Free
Format)

1
Figure 1: Single Column Resume

Figure 2: Double Column Resume

2
Figure 3: Free Format Resume (Image)

Structured Data (Name, Address, Email, phone_number, Skills [Hard Skills, Soft Skills],
Education, Experience, Languages)

Name

Address

Email

phone_number

Skills [Hard Skills, Soft Skills]

Education

Experience

The structured data comes in. json format which makes the HR department easy to read
the resume.

1.1. Problem Statement

People with diverse personalities come from a variety of fields and backgrounds. In the
same way, their resume writing style varies. They've worked on a variety of projects, and
each of them has unique writing style. As a result, each resume is unique.

3
Some people work in the human resources department. They will have to go through
hundreds of resumes on the internet. Executives summarize the resume, enter specific
information into their database, and then call the applicant for job counselling after
obtaining the resume. An executive spent around 10-15 minutes on each resume,
summarizing it and entering the information into the database. This project will help in the
automation of this procedure.

1.2. Implementation of module

By collecting the relevant information from the resume, a natural language processing
technique is employed to create a Hire ability system. Different NLP libraries, such as
NLTK and Spacy, are utilized for extraction. Natural language networks are usually taught
as unsupervised techniques, which implies that no previous tagging or labeling is done
before the model is trained.

Natural Language Processing is a branch of AI that integrates languages and computer

science to study the patterns and architecture of language. It helps to develop intelligent
systems that are based on machine learning and NLP algorithms which can read, analyze,
and extract meaning from text and voice.

1.2.1. Reason behind implementing NLP

The key reason I chose NLP for Hireability is that it can handle massive volumes of data in
seconds or minutes that would take days or weeks to analyze manually. NLP technologies
can instantly scale up or down to match demand, allowing us to have as much or as little
processing capacity as per requirement.
Aside from that, humans are prone to errors or may have personal biases that might distort
the findings while conducting repeated jobs like examining resumes one by one and other
textual data. In this case, NLP-powered solutions can be taught to understand company's
need and requirements in only a few steps. So, once they're up and going, they perform
better in terms of accuracy (Wolff, 2020).

4
1.2.2. Work Flow of NLP

Figure 4: Generic Workflow of NLP

Natural language processing comprises a wide variety of methods for analyzing human
language, based on machine learning techniques as well as rules-based and
computational approaches. Tokenization, lemmatization and stemming, parsing, part-of-
speech tagging, language identification are some basic NLP tasks. NLP tasks, in general,
break downs the language into smaller, essential components, attempt to comprehend
links between the pieces, and then examine how the components combine to form
meaning ( C, 2020).

The procedures given above represent the standard workflow for an NLP task.
The initial stage is generally text wrangling and pre-processing on the collection of
documents. Then there's parsing and some basic exploratory data analysis. The next
stage is to represent text using word embeddings and then do feature engineering.
Following that, we must select a model based on whether we are dealing with a
supervised or unsupervised learning scenario. The final step of any ML workflow is model
testing and deployment. The early processes of text pre-processing and EDA are covered
in ( 6.11.3. Data PreProcessing)

1.3. Aims

5
 • To take the help of the cutting-edge and latest NLP technology to enhance their
business processes.
• To extract the required information about candidates without having to go through
each resume manually.
• To replace slow and expensive human processing of resumes with extremely fast
and cost-effective software.

1.4. Objectives

• For information extraction, NLP model will be configured and reconfigured.

• A system for uploading resumes will be developed.
• The unstructured data will be transformed into structured form.

1.5. Artefacts

A subsystem is a fundamental component of a larger system. Dividing the system into

subsystems will aid in the development of a more advanced version. The following are the
three key artefacts that will be developed for this project in order to create a functional
model.

1. Admin management system

This sub-system is one of the most secretive aspects of the project. As it is a carefully
guarded part of the project, only the administrator can access using the admin email ID
and password. After the admin logged in, there'll be given the option to upload the
resume. Following the submission of the resume, there will have two sub-models:
resume parsing and information extraction.
The following is a brief list of how this artefact will be created and the tools required to do
it.
Methodology: SCRUM Design Tool: Figma
Logo Design: Figma
Frontend: HTML, CSS, JavaScript
Backend: Flask/Django
6
IDE: VsCode/ Sublime text Version
Control: Git
Browser: Google Chrome or Mozilla (Alternate)

2. File upload and Parsing system

This system is at the center of the overall project. In this sub system the admin will upload
the resume and go for further processing. The parser subsystem will use different libraries
to transform the submitted unstructured resume to a structured format. This will make it
much easier to examine, analyze, and grasp the data.
The following is a brief list of how this artefact will be created and the tools required to do
it.

Methodology: SCRUM
Data Requirement: dataset collected from Kaggle
Data analysis and evaluation: Jupyter Notebook/VsCode

IDE: VsCode/Sublime text

Version Control: Git

Browser: Google Chrome or Mozilla

Build Tools: Pip

3. Information extraction system

After parsing the data, we employ the NLTK, Spacy phrase matchers, regex to extract the
essential information. The required extracted data will be saved in JSON format.
Finally, the extracted data or dump JSON file is stored in a MySQL database for further
usage if it is needed.
The following is a brief list of how this artefact will be created and the tools required to do
it.
Methodology: SCRUM Design Tool: Figma Programming Language: Python 3
Technologies
Libraries: NLTK, Spacy
Data analysis and evaluation: Jupyter Notebook/VsCode

7
IDE: VsCode
Version Control: Git Build Tools: Pip

1.6. Scope and Limitations of the project

A purpose system can help in resolving the challenge of obtaining useful information from
a resume in a structured format. By resolving this issue, recruiters will be able to save
hours each day by eliminating manual resume screening. Bias in hiring is still prevalent,
thus this method may also address the bias hiring process and strengthen a non-bias
policy.

Purpose system is not able to solve the complex network issue such as:
• Excessive web traffic can significantly slow down or restrict access to a website
entirely. This occurs when the server receives more file requests than it can handle. •
Latency issues

1.7. Structure of the report

Introduction: It includes the description of purpose system, its types, problem domain
and project as the solution.
Literature Review: It includes the related project descriptions, algorithms used and the
diagram.
Answer to academic question: It includes answers to the academic question with the
proof.
Project methodology: It includes the methods taken to complete the project.
Technology and Tools: It includes the different types of tools and technology used
while building the system.
Artefact Designs: It includes the major artefacts of the system with its description and
diagrams.
Conclusion: It includes what has been discovered while writing and developing the
project.
8
 References and the Bibliography: It includes all the sources from where the project is
developed and report is written.

1.8. Academic Questions

• What would be the best approach to parse and extract useful information from a
resume in an effective manner using AI?
Explanation of academic question

Hiring an employee often causes complications if they hire based on a resume

since they have to check through all of the candidates' data such as education,
talents, job experiences, and so on. As a result, people may begin to wonder how
they would get through all of the resumes.They are always looking for a solution to
the problem. Machine learning or artificial intelligence (AI) could be used to extract
information from resumes as a solution.

2. Literature review
Agencies and different high-level companies have to deal with an extreme number of new
jobs seeking employees with different resumes. However, looking after those large
numbers of text data and filtering out the needed candidates is a burden on the brain and
more time consuming. Therefore, the essence of this literature review is on studying
resumes in different formats such as single-column resumes, double-column resumes with
extension.pdf,.docx, and how the suggested Information Extraction System converts that
unstructured information into structured layout through Parsing.

Accordingly, this review also helps to understand and apply several in-use and well
recognized algorithms currently being used in industries to reduce human labor.
Depending upon the Company's preference to hire employees, the Extraction System will
manage the gathered information with more readability and organized data forms.
Furthermore, the analysis of various machine learning algorithms and natural language
processing techniques would be equally carried out along with their proper implementation
and evaluation. The reviews from multiple research publications and journals are included
below.

9
2.1. End-to-End Resume Parsing and Finding Candidates for a Job Description
using

BRET

For evaluating the suitable candidates for various job openings in consideration to their
compatibility, the use of deep learning-based system has provided the end-to-end solution
for people seeking employment ( Bhatia, et al., 2019). They may be done in two stages:
first, by developing a resume parser that extracts all necessary information from candidate
resumes, and second, by conducting ranking using BERT phrase pair classification. The
BERT algorithm, which was used to classify sentence pairs, predicted the measure of the
correlation between both the job description and candidate profiles with 72.77 percent
accuracy. In this research, they explored the possibility of building a standard parser for all
types of resumes and determined that it was impossible to do so without losing information
in all situations, resulting in the unfair rejection of specific candidates' resumes. Instead,
they used LinkedIn-style resumes to scan without losing any information. They wanted to
investigate the vision-based site segmentation technique in the future in order to improve
structural comprehension of resumes. In addition, the study also creates a firm foundation
and a feasibility study that can lead to advancements in deep learning and language
representation being used in the hiring process. The system diagram below represents the
data flow and the completed task.

Figure 5: A system diagram showing data transmission and task completion.

10
2.2. Information Extraction from Free-Form CV Documents in Multiple Languages

The use of two natural language processing algorithms to extract important data from an
unstructured multilingual CV has provided a solution for selecting relevant document
parts and the similar particular information at the low hierarchy level (VUKADIN, et al.,
2021). It uses a machine learning method in NLP to obtain a high level of extraction
accuracy. In their practice, authors used the transformer architecture and its application
of the encoder part in the BERT language model. A dual model was developed to extract
both section and item level information from a CV document. A self-assessment model of
skill proficiency categorizes the retrieved Skills section from the dual model. The authors
claim that they have solved the CV parsing challenge by building an NLP system. The
recently introduced tokens [NEW LINE] and [SKILL] are shown to have trained to perform
as expected. In the future, adding additional CVs in other languages to the dataset would
improve performance in other languages.

Figure 6: The Information Extraction System from Free-form CVs: A High-Level Overview

11
2.3. Resume Information Extraction with a Novel Text Block Segmentation
Algorithm

Recruitment has evolved rapidly in the last decade, from traditional job fairs to web-based
recruitment platforms. As a result, (Wang & Zu, 2019) presented a resume parsing
pipeline that uses neural network-based classifiers and distributed embeddings from
beginning to end. The pipeline eliminates the laborious process of manually creating
several handcrafted elements. The proposed system combines text block segmentation
with resume fact identification using position-wise line information and integrated word
representations and named entity recognition using multiple sequence labeling classifiers
inside labeled text blocks. The ablation experiment was conducted by eliminating the CNN
layer from BLSTMCNN's-CRF and comparing various word embeddings testifying.
Personal information, job experience, education, project experience, publications, and
professional abilities are the six main areas they have focused on to normalize the resume
parsing process. Finally, the authors created an online resume parser based on the
proposed resume information extraction method, and it turned out that the system works
effectively in practice. In the future, they hope to improve the functionalities of online
resume parsers by embracing the ontology notion.

2.4. Information Extraction from Resume Documents in PDF Formats

( Chen, et al., 2016)focuses on the problem of extracting data from PDF-format resumes
and proposes a hierarchical extraction technique. Resume papers break a page into
blocks using heuristic criteria, categorize each block using a Conditional Random Field
(CRF) model, and approach the detailed information extraction problem as a sequence
labeling issue. According to the authors, the layout-based features have shown to be
especially beneficial for semi structured information extraction challenges, increasing the
average F1score by more than 20% in testing. In comparison to HTML resumes, PDF

12
resumes usually include more detailed information. They want to experiment with various
page segmentation techniques in the future to improve their understanding of the
document's layout and content. The report's method is explained below.

Figure 7: Workflow of Information Extraction from Resume

2.5. Study of Information Extraction in Resume

( Nguyen, et al., 2018) proposes Text Segmentation, Rule-based Named Entity

Recognition, Deep Neural Network Find Name Entities, and Text Normalization
approaches that automatically retrieve and process multiple resumes formats. To label the
sequence in the resume and then extract Name Entities from the labeled line, the author
defined the Deep Learning model as a mix of Convolutional Neural Networks, Bidirectional
Long Short-Term Memory, and Conditional Random Field. The developer collected
promising findings with over 81 percent F1 for NER when experimenting on a medium-
sized collection of CV information and compared this model to other systems. However,
there are some flaws in this system that might be addressed. The quantity of data to train
the model is insufficient. Thus, the more data, the more accurate the model will be. In
addition, the model's calculations necessitate the use of sophisticated computers. The
author wishes to improve the hardware systems to improve performance in the future.

13
2.6. Automatic Extraction of Usable Information from Unstructured Resumes to Aid

Search

( Kopparapu, 2015)proposes a natural language processing (NLP) system that focuses

on automated information extraction from resume to facilitate speedy resume search and
management for structured and unstructured resumes. According to them, it is a two-
pass method; in the first pass, the resume is divided into a group of successive labeled
blocks that indicate the broad data that the block includes. Then, in the second pass,
information is retrieved in detail. They applied a variety of heuristics and pattern
matching algorithms for the extraction procedure. According to the results of experiments
conducted on many resumes, the suggested system can grip a wide range of resumes in
various document formats with an accuracy of 91% and a recall of 88%. In the future,
they expect to enhance the accuracy with which a CV is picked for high-skilled
employment. The data flow and completed tasks are represented in the system diagram
below.

Figure 8: Workflow of the System

2.7. Analysis and Findings

The research and execution of several algorithms to extract the legitimate number of
applicants based on their resumes has been raised following the detailed analysis of the
proposed literature review. Based on the studies and research papers, it has been seen
that the implementation of various algorithms like Bidirectional Encoder Representations

14
from Transformers (Bert) Model, Neural Network, NLP have been fruitful for information
extraction. BERT is a language representation pre-training approach that involves training
a general-purpose "language understanding" model on a considerable text corpus, such
as Wikipedia, and then using that model to downstream NLP tasks. Because it is
profoundly bidirectional and the first unsupervised system for pre-training NLP, the BERT
technique beats RNN, LSTM, and Bi-LSTM results. BERT was trained unsupervised with
just a plain text corpus, which is noteworthy because a massive quantity of simple text
corpus is significant because a large amount of primary text data is publicly accessible on
the web in different types of languages.

Even though the research makes extensive use of different approaches, extracting
information without losing candidate information from resumes is seen to be a challenging
task that requires complex analysis. However, compared to other methodologies
described in the papers, the NLP method is simple to adopt and provides meaningful
results. It focuses on allowing computers to understand natural language content in a
human-like way and develop intelligent systems capable of comprehending, evaluating,
and retrieving information from the text. We can also see the ratio of correct predictions
made against the size of the test data. The recall and f1 scores are likewise good in terms
of overall comparison. Its capability also astounds the researcher since it effectively
extracts relevant information from unstructured resumes. So, after evaluating all of the
data, I also discovered how NLP is used to extract useful information from resumes. On
the other hand, Cosine similarity will measure the score based on cv and job description.
After performing the cosine, similarity recall is measured. The score based on high
probability will quickly help in the job match.

2.8. Conclusion of literature review

Modern science has developed an excellent learning and working environment in the
twenty-first century. Likewise, AI-based solutions are gaining worldwide recognition and
implementation for solving the challenge of filtering necessary applicants for a company.
As a result, different information extraction systems are created in the present time,
replacing the traditional and time-consuming one-by-one approach with a simple filtering
out of what is required for the Company's post.

The reviewed literature addressed a wide range of information extraction concepts and
research-based procedures and their main approaches. The majority of the research
15
material uses a combination of ML and DL-based methods. After analyzing all of the
review findings, it is easier to conclude that extracting meaningful information from
resumes using NLP and its different techniques like Regex and Spacy simplifies the
recruiting process significantly minimizes time complexity in a larger way

2.10. Competitive analysis

The two applications listed below have been evaluated for doing competitive analysis as
well as studying the system and services they deliver.

1. Skillate

In general, Skillate helps in the optimization of the complete recruiting value chain, from
job request creation through resume matching and applicant participation. Deep
Learning is used by the system to extract information from even the most complex
resumes. This application allows users to upload resumes in a various format,
including .PDF, DOC, DOCX, and ZIP. The most complicated resumes are processed in
1-3 seconds by the AI enabled parser.
The system is a powerful decision-making tool that makes recruiting simple, quick, and
fair. When it comes to the system's drawbacks, the initial login to the home page takes
time. There could be features that cater to various areas of the hiring process. The ability
to deliver answers in multiple nations and languages is also essential in features that are
not available in the system.

16
Figure 9: Skillate Software

2. DaXtra

This program continues the same pattern as the last one, quickly, effectively, and
accurately extracting information from CVs. The organized data generated from the
system follows a well-defined XML or JSON format. Personal and contact information,
work history, and education background are all included in CV data. Another component
is the Native talents taxonomy, including industry skills, job titles, and academic and
professional credentials. The system also supports custom taxonomies. One notable
feature of this system is it utilizes multilingual text-understanding technologies. This
provides worldwide users with a thorough understanding of the geographical name,
address, and number forms across a variety of document types.
When it comes to the system's disadvantages, there are a lot of bells and whistles, which
might be confusing for non-technical recruits. The searches aren't very good, and because
each search platform uses various search inputs, a single search entry for many platforms
isn't very useful.

17
Figure 10: DaXtra Software

3. Project Methodology

A collection of guiding concepts and practices for planning, managing, and executing
projects is known as project management methodology. The project management
approach that we select affects how work is prioritized and finished. Here, are a few well-
known project management approaches in use today.

1. Waterfall methodology
2. Agile methodology
3. Kanban methodology

Agile is a software development process that is continuous and responsive. High levels of
communication and cooperation, rapid and convenient reactions to change, adaptable
planning, and continuous improvement are all characteristics of Agile development.
Scrum, Kanban are just a couple of the techniques and sub-frameworks that have resulted

18
in the emergence of the agile project management concept. Each framework has its own
set of advantages and disadvantages.

3.1 Scrum Methodology

To do Hire ability project as yearlong project I have Chosen scrum. It is the most widely
used Agile methods framework due to its ease of use and efficiency.

3.1.1. Reason behind Selection

The major reason I chose this method is:

• It is a transparent approach to project management.

• The tasks and the progress of each task are constantly visible during the sprint.
• It is easy to learn and use.
• It even reduces time taken to get product to market.

Scrum divides work into small cycles known as "sprints," which typically run approximately
one to four weeks. For each sprint iteration, work is taken from the backlog. For the course
of the sprint, small teams are supervised by a Scrum Master, after which they analyze
their work in a "sprint retrospective" and make any required changes before beginning the
next sprint. Testing thoroughly, paying close attention to detail, and learning from prior
Sprints all contribute to a high level of quality (teamwork, 2021) .

Figure 11: Sprint Backlog

3.2. Initial Gantt Chart

19
Figure 12: Initial Gantt Chart
3.3 Additional info

3.3.1 Resources

The resources necessary for the entire project's development are listed below.

Hardware Requirements:
• PC with 128 GB of SSD, Python Development Environment and a stable
internet connection
• Windows and Linux for testing

• Figma for UI/UX design

20
Software Requirements:

• Figma for logo design

• Google Chrome or Mozilla (Alternate) for research and debugging

• Visual Studio Code for Python Development

• Git for version control

• Google Drive for the document (Word/ PDF) controls

• Microsoft Visio/draw.io for creating WBS and other diagrams

• Excel for creating Gantt Chart

• MySQL for storing data if needed

• Pip for installing python packages

• NLTK and SpaCy for resume parsing and extraction

• Flask for developing web application

• Jupyter Notebook or Vscode for evaluating and exploring data

3.3.2 Client

Mr. Sarjil Napit would act as the project's supervisor, while Mr. Yogesh Bikram Shah, a
teacher at Herald College Kathmandu, would act as the project's reader.

4. Technology and Tools

During the project's development, a variety of tools were used. The list of Software tools
used in the project is divided into two categories: hardware and software tools.

1. Programming Language:
For this project, I employed the Python programming language. The major reason for
using Python is because of its simple syntax, which makes it versatile, simple to use, and
quick to develop. Another reason to prefer Python is that it can be used for web
development and is ideal for ML and AI applications.

2. Framework:
When I began exploring frameworks, Django was at the top of the list. Flask, Laravel,
Ruby on Rails, and CakePHP were among the other alternatives, but according to

21
research and evaluations, the Django framework was more reliable, fast-simple, and, most
importantly, perfect for any web application project. It's a Python web frame. As a result, I
began to focus on Django for the project's backend by watching videos, reading manuals,
and so on.

3. Web Technologies:
For frontend development, HTML, CSS, JavaScript, and Bootstrap were chosen. HTML
provides the foundation for a website's structure, which is then improved and updated by
CSS and JavaScript. CSS is used to manage the appearance, style, and layout of a page,
whereas JavaScript is used to control the behavior of certain components.

4. Design Tool:
The design is the most key component of any project. So, I used Figma for the various
design sections since it is a free graphic design application that is simple to use. Another
advantage of using Figma is we can create, produce, and export files very quickly. I've
also used draw.io, a free-to-use online tool, for additional diagrams like activity and use-
case. It also comes with lots of Templates for creating simple diagrams and flowcharts.

5. IDE:
For the coding side, I used vscode. As, VS Coding is a highly capable code editor with
built in git integration. It attracted to me since it is simple to use, write code in, add files,
and most importantly, it provides syntax highlighting and bracket-matching.

6. Version Control:
When working on any project, using Version Control is essential since it allows individuals
to rapidly track changes made to the project and ensures that everyone is working on the
most recent version. So, using a variety of commands, I uploaded my project to Github
and committed it every time I made a change to the code.

7. Browser:
I choose Google Chrome as my browser since it is simple to use and is faster. It also
loads web pages with a single click and can open many tabs at once.

22
5. Artefact Design

5.1. System Requirement Specification

5.2. Functional Decomposition Diagram

A chart called FDD is used to represent the separate components of the method and their
varied levels relationships to one another. The diagram depicts a technique in a top-down
order. It has a rectangle for the function name and a connecting arrow.

Figure 13: Functional Decomposition Diagram

5.3. Legend

23
5.4. Admin Management System (AMS)

Requirement- Code Requirement Specification MOSCOW

AMS-F-1 The admin should be able to click the Must Have

Login button on the system.

AMS-NF-1.1 Login form should be opened. Must Have

AMS-F-2 System should allow admin to login. Must Have

AMS-F-2.1 Email should be valid.

Must Have
AMS-F-2.2 Password should be valid. Must Have

AMS-F-2.3 For a successful login, the admin must Must Have

have a valid username and password.

AMS-F-2.4 The entered username and password must Must Have

be validated by the system.

AMS-UR-2.1 If the login and password are incorrect or Must Have

empty, the system should produce an
error message.

24
6. Diagrams

6.1. Activity Diagram

The behavior of a system is depicted in an activity diagram. The control flow from a start
point to a completion point is represented in an activity diagram, which shows the
numerous decision routes that exist while the activity is being performed. The activity
diagram for the admin management system is shown below (geeksforgeeks, 2018).

Figure 14: Activity Diagram of admin management system

25
Figure 15 activity diagram of parsing system

6.2. Use Case Diagram

The interactions between the system and its actors are mostly identified with the use case
diagram. In use-case diagrams, use cases and actors define what the system does and
how the actors interact with it. There is just one user in the diagram below, and that is the
admin. Admin can log in, navigate to the upload page, upload their resume, and view the
extracted data (ibm, 2021).

26
 Admin

Figure 16: Use Case Diagram

6.3. Sequence Diagram

A sequence diagram displays the order in which items interact in a sequential way. It
shows how and in what sequence the components of a system work together. The
sequence diagram for the admin management system is shown below (geeksforgeeks,
2018).

27
Figure 17: Admin Management System

Figure 18: File Upload System

28
6.4. Wireframe Designs

Many software programs, such as Balsamiq, Figma, and in vision, have been designed to
build wireframes. Figma was my tool of choice. According to the website's requirements, I
created a wireframe that shows the website's simple layout, which includes a navigation
bar, content section, and footer. Wireframes do not contain any of the system's logic.
Below is the home page of the application.

6.4.1. Home Page

Below is the login page of application.

29
6.4.2. Login Page

Below is the parsing page of application.

6.4.3. Parsing Page

30
Below is the information Extraction Page of the System.

6.4.4. Information Extraction Page

31
Below is the FAQ Page of the system.

32
6.4.5. FAQ Page

Below is the terms and condition page of the system.

33
6.4.6. Terms and Condition Page

6.5. Design Process

The goal of design is to get as near as possible to the final design. As a result, I built a
design that featured all of the pages' important content by picking the appropriate colors,

34
font, and structure. All of the visual material, including images, icons, and buttons has
been created. I reduced the number of unnecessary colors that might cause users to
become distracted. The design was made with Figma. My main considerations have been
usability, layout, and consistency. The design has provided a general notion of what the
upcoming product would look like.

Figure 19: Design Part

6.6. Development

The final and most essential step is development. I built a functional website in
development using the graphic components that were created during the wireframe. The
home page is created initially, followed by all sub-pages. The static web page elements
35
modeled during the mock-up were generated. The development languages used were
HTML, CSS, JavaScript, Bootstrap, and Flask. The web pages are user-friendly and
responsive, with plenty of visuals on screen. To make the website more engaging, various
accessibility options have also been included.

6.7. Product Backlog

As a I want to So that Acceptance Priority

Criteria

Admin Login the I can upload the Login credentials Must Have
system resume

Admin Select files I can parse the A system to open Must Have
from the resume. Folder
Folder

Admin See the I can choose the Parse the resume Must Have
extracted right candidate as
information. per the need.

Admin Reupload an I can reupload A system to open Must Have

file from the any Files to parse Folder
files the information.

6.8. Accessibility

The most significant factor in determining a site's worth is its accessibility. It is important
that the Internet be accessible to all in order to provide those with disabilities with equal
access and opportunities. People with impairments can also benefit from a web that is
accessible. Two accessibility symbols have been used on these websites.

36
a) Invert Color:
Inverting the color helps to boost visibility without raising the screen's brightness when the
screen is in direct sunlight. Another significant benefit is that it helps those who are
colorblind or suffer from night blindness to use their gadget without becoming anxious.

Figure 20: Invert Color

b) Increasing and decreasing the text size:

Many persons with low vision do not use magnifying software and are unaware of how to
change the font size on their browser. Larger fonts may also attract to certain persons
with cognitive impairments. To use this technique, there are buttons that enable users to
gradually increase, decrease, or keep the text size of all the text on the page at the same
level.

37
 Figure 21: Increasing and Decreasing Text size

6.9. System Work Flow

Figure 22: System Work Flow

6.10. Testing and Evaluation

6.10.1. Accessibility Testing

A) Lighthouse
To do my testing, I employed a range of tools. Lighthouse is one of them. Lighthouse is a
free open source, automated web page quality optimization tool that is mostly used for
accessibility testing. Lighthouse provides the Chrome DevTools Audits panel. To run a
study, we must do the following:
• First, we'll need to download Google Chrome for Desktop, and then we'll need to open
Chrome DevTools.
• We'll need to choose inspect.
38
• DevTools provides a list of audit groups after selecting Lighthouse. We may look at many
options based on our preferences, but accessibility is a must.
• Select "Generate Report" from the drop-down menu. After 30 to 60 seconds, Lighthouse
provides a page report.
The lighthouse then shows us the error fields, which we can utilize to fix our mistakes.

Figure 23: Light House Testing

B) Color Contrast Checker

In Figma, I tested the design element for the second time. Color contrast is a key factor in
making the site more accessible to everyone. To make writing simpler to read, the contrast
between the text and the context should be significant. So, prior to the development
phase, I merely evaluated and fixed all of the color contrast in the design process. To
double-check:

• First, we must select plugins.

• Then, choose A11y-Color Contrast Checker.
• Finally, press the check button. And, if there's a color issue, it'll show up on the board,
and we'll be able to fix it from there.

39
 Figure 24: Color Contrast Test

6.10.2. Black Box Testing

Testing must be done once the project has been completed to ensure that the app
functions properly. The test runs under various conditions and for each activity developed
for the system. The test that will be carried out is known as unit testing. The Black Box's
major goal is to determine whether the software is performing as expected in the
requirement document and whether it is satisfying user expectations.

a) Login Activity

Test Case No. 1

Test case To test whether the application successfully allow login for admin.
Objective
Test Data: Test Data Expected Result Actual Result

40
 i. Username:
Hireability ii.
Password: Login Successfully Login Successfully
Hireability#123
iii. Email:
[email protected]

Test Results The above result indicates that login of admin in the application
works well.

Screenshot of test 1:

Figure 25 : Screenshot of Test 1

Test Case No. 2

Test case To test whether the application pass invalid login and left attempt
Objective message after invalid input.

Test Data: Test Data Expected Result Actual Result

41
 i. Username: Hire
ii. Password:
Hireability#123 Left Attempt Left Attempt
iii. Email:
ps@

Test Results The above result indicates that login of admin in the application
works well.

Screenshot of test 2:

Figure 26 : Screenshot of test 2

Test Case No. 3

Test case To test whether the admin is able to upload File from the File
Objective explorer application.

42
 Test Data: Test Data Expected Result Actual Result

Click Upload Button Open file explorer Open file explorer

and upload file and upload file and upload file

Test Results The above result indicates that the admin is able click upload button
and upload files.

Screenshot of test 3:

Figure 27: Screenshot of test 3

43
 Figure 28: Screenshot of test 3

Test Case No. 4

Test case To test whether the necessary information from the resume is
Objective parsed.

44
 Test Data Expected Result Actual Result

Shown name, email,

Test Data:
Show name, email, mobile number,
Upload the resume mobile number, technical
technical skills, soft skills, soft skills,
skills, education, education,
Language, Language,
experience, address experience, address

Test Results The above result indicates that the necessary information from the
resume has been parsed.

Screenshot of test 4:

Figure 29: Screenshot of test 4

45
6.11. Module Implementation

This part contains information about data collecting, how the code is working in the project,
different libraries and modules utilized in the project, and so forth.

6.11.1 Important Necessary Libraries and Module

In today's world, everyone wants to complete the tasks as early as possible, but it all relies
on the methods they use. There may be a point when we must write many lines of code; if
we continue writing code on our own, this will take time. A Library is a group of
interconnected modules. It provides code bundles that may be reused in a variety of
programs. Modules also play a role in Python. We may define commonly used functions
as modules and import them into a code wherever there is a necessity, rather than
repeating the same code in multiple programs and making the code complicated. The
following are some of the most important libraries and modules:

1. Pandas
Pandas is a Python library which is an open source. It's a tool for analyzing data.
2. OS
In Python, the OS module has methods for creating and deleting folders, retrieving
their contents, altering and identifying the current directory, and so on.

3. IO

46
 The io module is used to control file-related input and output activities. The
advantage of utilizing the IO module is that we can extend the capability to allow
writing to Unicode data using the classes and methods provided (journaldev, 2022).

4. Utils
Python Utils is a collection of simple Python methods and classes that simplify and
shorten common patterns.

5. Spacy
SpaCy is an open-source package library for advanced natural language
processing implemented in Python. SpaCy has pre-trained pipelines with
tokenization support.

6. pprint
The pprint module allows to "pretty-print" any python data structure in a more
understandable and well-formatted fashion.

7. Matcher
The Matcher assists in the finding of words and phrases by utilizing rules that
describe their token properties. After applying the matcher on a Doc, we can see
the matched tokens in context.

8. Multiprocessing
In Python, multiprocessing is a built-in package that allows the system to execute
many processes at the same time.

9. Warnings
Warning messages are shown using the warn () method from the 'warning' module.

10. Nltk.corus
The modules in this package provide functions for reading corpus files in a variety
of formats. A corpus is simply a set of texts that are used as input.

11. Re

47
 A RegEx, also known as a Regular Expression, is a string of characters that defines
a search pattern. This module's functions allow to see if a given string matches a
given regular expression.

12. NLTK
NLTK is a Python toolbox for working with natural language processing. It gives us
access to a number of text processing libraries as well as a large number of test
datasets.

13. Docx2txt
A pure Python-based library for extracting text from docx files.

14. Constants
The module is just a separate file that contains variables, functions, and other data
that is imported into the main file.

15. String
This module will allow to retrieve string constants easily. Cap words is a single utility
function in the Python string module.

16. Pdfminer
PDFMiner is a tool that extracts text from PDF files. It focuses completely on
gathering and processing text data, unlike other PDF-related tools.

6.11.2. Data Collection

Because training data is not necessary, I only need testing data for my "Hireability"
system. So, I got some data from Kaggle's "Hire A Perfect Machine Learning Engineer"
and the rest through beam jobs. There were 48 resume templates available in the beam
jobs collection.

6.11.3. Data Pre-Processing

48
Natural Language Processing is a subfield of data science that works with textual data.
When it comes to handling the Human language, textual data is one of the most
unstructured types of data available. NLP is a technique that operates behind the it,
allowing for extensive text preparation prior to any output. Before using the data for
analysis in any Machine Learning work, it's critical to analyze the data. To deal with NLP-
based problems, a variety of libraries and algorithms are employed. For text cleaning, a
regular expression(re) is the most often used library. The next libraries are NLTK (Natural
language toolkit) and spacy, which are used to execute natural language tasks like
eliminating stopwords.
Preprocessing data is a difficult task. Text preprocessing is done in order to prepare the
text data for model creation. It is the initial stage of any NLP project (Deepanshi, 2021).
The following are some of the preprocessing steps:
• Removing Stop words
• Lower casing
• Tokenization
• Lemmatization

1. Tokenization

The initial stage in text analysis is tokenization. It enables to determine the text's core
components. Tokens are the fundamental units. Tokenization is beneficial since it divides
a text into smaller chunks. Internally, spaCy determines if a "." is a punctuation and
separates it into tokens, or whether it is part of an abbreviation like as "B.A." and does not
separate it. Based on the problem, we may utilize sentence tokenization or word
tokenization.
a. Sentence tokenization: using the sent_tokenize () function, dividing a paragraph into
a collection of sentences.
b. Word tokenization: using the word_tokenize () technique, dividing a statement into a
list of words.

49
 Figure 30: Tokenization

For the Token class, spaCy gives a number of characteristics ( chavan, 2020). The
following is the most frequently applied attribute:
• is_stop determines whether the token is a stop word or not

2. Removing Stop words

To eliminate noise from data, data cleaning is essential in NLP. Stop words are the most
frequently repeated words in a text that give no useful information. The NLTK library
includes a list of terms that are considered stop words in English. [I, no, nor, me, mine,
myself, some, such we, our, you'd, your, he, ours, ourselves, yours, yourself, yourselves,
you, you're, you've, you'll, most, other] are only a few of them.

The NLTK library is a popular library for removing stop words, and it eliminates about 180
stop words. For certain difficulties, we can develop a customized set of stop words. Using
the add technique, we can easily add any new word to a collection of terms.

Figure 31: Stop words

3. Lemmatization
The process of reducing inflected forms of a word while verifying that the reduced form
matches to the language is known as lemmatization. A lemma is a simplified version or
base word. Lemmatization uses a pre-defined dictionary to saves word context and verify
the word in the dictionary as it decreases. Organizes, organized, and organizing, for
example, are all forms of organize. The lemma in this case is organize. The inflection of a
word can be used to communicate grammatical categories such as tense (organized vs

50
organize). Lemmatization is required since it aids in the reduction of a word's inflected
forms into a particular element for analysis. It can also assist in text normalization and the
avoidance of duplicate words with similar meanings ( Agrawal , 2021).

Figure 32: Lemmatization

4. Lower casing

When the text is in the same case, a computer can easily read the words since the
machine treats lower and upper case differently. Words like Cat and cat, for example, are
processed differently by machines. To prevent such issues, we must make the word in the
same case, with lower case being the most preferable instance. In python lower () is a
function that is mostly used to handle strings. The lower () function accepts no parameters.
It converts each capital letter to lowercase to produce lowercased strings from the
provided string. If the supplied string has no capital characters, it returns the exact string.

Figure 33: Lower Casing

6.11.4. Code and its explanation

At first, I have imported all the necessary Libraries and modules.

51
The first step after importing the libraries is to read the resume. As resumes do not have a
set file type, they can be in any format, including.pdf,.doc, and.docx. As a result, our key
priority is to read the resume and turn it into plain text. I used pdfminer and doc2text
Python modules for this. Text can be extracted from.pdf,.doc, and.docx files using these
modules.

a) Extract the text from pdf

52
 Figure 34: Extract the text from pdf

Here, I have defined a function to extract the plain text from .pdf files. The pdf_path
parameter gives path to pdf files that is to extracted. We will begin by establishing a
resource_manager instance. Then, using Python's io module, we will then build a file-like
object. The next step is to build a converter. We'll use the TextConverter for that. Finally,
we construct a PDF interpreter object that will extract the text from our resource
management and converter objects. It returns the iterator of string of extracted text.

b) Extract Text from Doc files

Figure 35: Extract Text from Doc files

Here, I have defined a function to extract the plain text from .docx or .doc files. The
doc_path parameter gives path to .doc, .docx files that is to extracted. It returns the string
of extracted text.

c) Detect the file extension

Figure 36: Detect the file extension

Here, I have defined a function to detect the file extension and call the text extraction
function accordingly. If the file is on .pdf or .docx or .doc then it returns the text. The
53
file_path parameter gives path of the file of which text is to be extracted. After that the
param_extension is the extension of the file ‘file_name’.

d) Extract the entities

Figure 37: Extract the entities

Here, I have defined a function to extract all the raw text from the sections of resume. I
have already created the list for sections of resume in constant.py file. The split () function
divides a string by the separator supplied and produces a list object containing string
elements. Whitespace character, such as, \t, \n, are used as the separator by default. The
text parameter gives the raw text of the resume. It returns the dictionary of the entities.

54
 e) Extract the email

Figure 38: Extract the email

Here, I have defined a function to extract the email id from the text. The text parameter
gives the plain text extracted from the resume files. As we know, email addresses have a
set format, which includes an alphanumeric string, a @ symbol, another string, a. (dot),
and a string at the last. We can extract such expressions from text using regular
expressions.

f) Extract Name from Resume

55
 Figure 39: Extract Name from Resume

Here, I have defined a function to extract the name from the text. The text parameter gives
the plain text extracted from the resume files. Whereas the matcher parameter is the
object of spacy.matcher. Spacy is a text and language processing module. It includes pre-
trained models for tagging, parsing, and entity recognition. The primary goal here is to
extract names using Entity Recognition. One of the phases in extracting information from
unstructured text is rule-based matching. It recognizes and extracts tokens and words
based on patterns. Regular expressions can be used in rule-based matching to extract
entities. We can extract a first and last name, which are generally proper nouns, using
rule-based matching. A pattern is a collection of items that define the set of tokens to be
matched. It has two PROPN POS tags in it (proper noun). As a result, the pattern is made
up of two objects, each with PROPN POS tags. The NAME and the match id are then
used to add the pattern to Matcher. Finally, matches are found using their starting and
ending indexes.

g) Extract Mobile Number

56
 Figure 40: Extract Mobile Number

Here, I have defined a function to extract the mobile number from the text. Regular
expressions will be used to retrieve phone numbers. Phone numbers come in a variety of
formats, including (+91) 1234567890, +911234567890, +91 123 456 7890, or +91
1234567890. As a result, I created a generic regular expression which can match all
identical phone number patterns.

h) Extract Skills from the resume

Figure 41: Extract Skills from the resume

After gathering some basic information about the individual, let's focus on what matters
most to recruiters: skills. For that, I've created a function that extracts skills from spacy
text. The object of spacy.tokens is the NLP text parameter. Noun chunks are extracted
from the text using the noun chunks parameter.

57
A method known as tokenization can be used to extract skills. In order to perform
tokenization, we need to establish a sample dataset against which we can compare the
abilities in a given résumé. For this, I have created a comma separated values (.csv) file
with the necessary skill sets. For instance, if I'm in HR and need someone with NLP, ML,
and AI abilities, I can create a csv file with the following keywords: machine learning,
artificial intelligence, natural language processing. Following that, we renamed the file to
skills.csv
Then, we can proceed to tokenize our extracted text and compare the skills to those in
skills.csv. The panda’s module will be used to read a csv file. After reviewing the material,
we will remove all of the stop words from our resume. At last, it returns the list of skills
extracted.

i) Extract Technical Skills

Figure 42: Extract Technical Skills

Non-technical skills are extracted using the same procedure as technical skills.

j) Extract Education and Year

58
 Figure 43: Extract Education and Year

The degree and the year of graduation will be the details we will extract. If XYZ finished BS
in 2022, for example, we will extract a tuple like ('BS', '2022'). We'll need to get rid of all
the stop words for this. We'll use the nltk module to load a long list of stopwords, which
we'll then remove from our resume text. Hr teams are quite specific about the level of
education or degree necessary for a given job. As a result, we'll be putting together an
EDUCATION list that will include all of the similar degrees that meet the standards.
k) Extracting address

Figure 44: Extracting address

Here, I have defined a function to extract the address from the text using Spacy.
GPE often refers to geopolitical entities such as cities, states, nations, continents,
59
 and so on. In addition, LOCATION also can symbolize well-known mountains,
rivers, and other natural features. The act of identifying named entities in
unstructured text and then categorizing them into pre-defined categories is known
as Named entity recognition (NER).

l) Extract Experience

Figure 45: Extract Experience

Here, I have defined a function to extract the experience from the resume text. The
resume_text parameter gives the plain text extracted from the resume files. After
doing different preprocessing part. It searches the word ‘experience’ in the chunk
and then print outs the text after it. At last, it returns the list of experience.

6.11.5. Output

60
 Figure 46: Output

.json format is used to display the retrieved data. This allows the HR staff to quickly identify
candidates that meet the organization's needs. This technique also benefits the HR
department by saving time by eliminating the need to manually review each resume.

7. Conclusion
A normal resume is a compilation of information about a person's work experience,
academic background, qualifications, and personal details. These elements might be
present in a variety of ways or not at all. It's difficult to keep up with the jargon used in
resumes. A resume is made up of corporate names, institutions, degrees, and other
information that can be written in a variety of ways. It will take time to review all the resume
by an individual.

Machine works faster than human and their accuracy to do any task was also good.
Therefore, I have made a system which includes machine learning that extract the
important information from resumes within a minute or less than a minute. The hiring
individual can use this system for hiring any individual.

8. Critical Evaluation

8.1. Final Report

While working on my report, I experienced several issues. The primary issue was that I got
tiny bit confused throughout the testing part. The diagrams and grouping of some topics
into subtopics were the next difficult phase. However, after discussing with the teacher, I
was able to resolve this issue. Aside from the flaws, the positive aspect of the report is that
it is based on facts, researched information, and legitimate proofs.

8.2. Finding and process

There was some uncertainty when it came to select the right libraries and modules for data
extraction. After conducting research, I was able to select a library. Following that,
obtaining the appropriate dataset was difficult. I looked for a number of resumes dataset

61
but couldn't find one in the correct format. After that, I went online and found several
templates and started using them for training data.
8.3. System

While developing the system, there were a number of technical difficulties. Following the
development of the module, the next step is to integrate it into the system. I was having
trouble integrating the module while working on the flask. I tried everything but couldn't get
the component to work. As a result, I switched to Django as my framework. I was able to
integrate after reading through several python and Django documentation. The integration
part took up more time than expected.

Despite the time constraints, the majority of the features proposed in the proposal were
implemented. However, the Tika parser library, which I specified in the proposal for the
information extraction section, has not been implemented. It is used to detect the kind of
document. As I began working on the project, I found that Spacy and Nltk were more
suitable to my requirements, so I chose them over Tika.

8.4. Future Work

As stated in the paper, the project has a broad reach in the current context. The proposal's
majority of proposed features have been implemented. So, if I continue working on this
project, I intend to create a database for the system where the admin may keep the
extracted data. Further, future study will include a more in-depth examination of certain
techniques, further research on other libraries, and new approaches to explore different
methods.

8.5. Self-reflection

After working on a project, everyone learns something new. After completing this project, I
learned several new things about information extraction. I learnt how to deal with various
NLP libraries, how to implement the module, and so on. Aside than that, I learnt about
Django by watching videos and reading documentation. I also learned how to integrate the
system from this portion.

62
9. Evidence of Project Management

9.1. Gantt Chart

63
 Figure 47: Final Gantt Chart

9.2. Project Management tool

Figure 48: Project Management

10. References
• C, B. P., 2020. towardsai. [Online]
Available at: https://towardsai.net/p/nlp/natural-language-processing-concepts-
andworkflow-48083d2e3ce7 [Accessed 2022].
• Agrawal , R., 2021. analyticsvidhya. [Online]
Available at: https://www.analyticsvidhya.com/blog/2021/06/must-knowntechniques-for-
text-preprocessing-innlp/#:~:text=Text%20preprocessing%20is%20a%20method,text
%20in%20a%20diff erent%20case. [Accessed 2022].
• Bhatia, V., Rawat, P., Kumar, A. & Shah, R. R., 2019. End-to-End Resume Parsing
and Finding Candidates for a Job Description using BERT. arxiv.
• chavan, J., 2020. medium. [Online]
Available at: https://medium.com/@jeevanchavan143/nlp-tokenization-
stemminglemmatization-bag-of-words-tf-idf-pos-7650f83c60be [Accessed 2022].
• Chen, J., Gao, . L. & Tang, Z., 2016. Information Extraction from Resume
Documents in PDF. researchgate.
64
• Kopparapu, S. . K., 2015. Automatic Extraction of Usable Information from
Unstructured Resumes to Aid Search. ieeexplore.
• Nguyen, V. V., Pham, V. . L. & Vu, N. . S., 2018. Study of Information Extraction in
Resume. semanticscholar.
• D., 2021. analyticsvidhya. [Online]
Available at: https://www.analyticsvidhya.com/blog/2021/06/text-preprocessing-
innlp-with-python-codes/ [Accessed 2022].
• geeksforgeeks, 2018. geeksforgeeks. [Online]
Available at: https://www.geeksforgeeks.org/unified-modeling-language-
umlsequence-diagrams/ [Accessed 2022].
• geeksforgeeks, 2018. geeksforgeeks. [Online]
Available at: https://www.geeksforgeeks.org/unified-modeling-language-uml-
activitydiagrams/#:~:text=An%20activity%20diagram%20is%20a,the%20activity%20is
%20 being%20executed.
[Accessed 2022].
• ibm, 2021. ibm. [Online]
Available at: https://www.ibm.com/docs/en/rational-soft-arch/9.6.1?topic=diagramsuse-
case
[Accessed 2022].
• journaldev, 2022. Python IO Module. journaldev.
• Kurama, V., 2021. nanonets. [Online]
Available at: https://nanonets.com/blog/information-extraction/ [Accessed
2022].
• teamwork, 2021. teamwork. [Online]
Available at:
https://www.teamwork.com/project-management-guide/projectmanagement
-methodologies/ [Accessed 2022].
• VUKADIN, D., KURDIJA, A. . S., DELAČ, G. & ŠILIĆ, M., 2021. Information
Extraction From Free-Form. ieeexplore.
• Wang, X. & Zu, S., 2019. RESUME INFORMATION EXTRACTION WITH ANOVEL
TEXT BLOCK SEGMENTATION ALGORITHM. researchgate, Volume 8.
• Wolff, R., 2020. monkeylearn. [Online]
Available at: https://monkeylearn.com/blog/nlp-benefits/ [Accessed
2022].

65
66