www.ijcrt.

org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882

JOB RECOMMENDATION SYSTEM

BASED ON SKILL SETS
G.Mahalakshmi1, A.Arun Kumar2, , B.Senthilnayaki1, J.Duraimurugan1
1
Teaching Fellow, Department of Information Science and Technology, Anna University Chennai, India
2
MCA student, Department of Information Science and Technology, Anna University Chennai, India

Abstract Machine learning is a sub-field of data science that concentrates on designing algorithms that can learn from and make
predictions on the data. Presently recommendation frameworks are utilized to take care of the issue of the overwhelming amount
of information in every domain and enable the clients to concentrate on information that is significant to their area of interest.
One domain where such recommender systems can play a significant role to help college graduates to fulfill their dreams by
recommending a job based on their skill set. Currently, there are plenty of websites that provide heaps of information regarding
employment opportunities, but this task is extremely tedious for students as they need to go through large amounts of
information to find the ideal job. And many students are not aware of which job is suitable for them. Nowadays, the IT fields
are in a boom. Many engineering students are learning some technical skills by doing some courses but they don’t know which
skill is for which job. Simultaneously, existing job recommendation systems only take into consideration the domain in which
the user is interested while ignoring their profile and skillset, which can help recommend jobs that are tailor-made for the user.
This paper examines the user’s resume then compares the knowledge of degree, soft skills, hard skills, and the projects he has
done and then only the system recommends the jobs for that user. The system not only recommends the jobs but also shows
the score of his/her resume for the respective job. Then, the system also recommends skills to improve the scores of their
resume.

Index Terms: Skill set, Cosine Similarity, Jaccard Similarity, Euclidean Similarity

I. INTRODUCTION

A recent report claims that most college graduates have difficulty in choosing their domain in their job. Many engineers are
trying to shift the domain from their field to IT. So, they are doing some courses in online and randomly searching for a job.
Nowadays, IT fields are the targets of many students but they don’t know which domain is fit for them. To avoid this situation
candidates, need a Job recommendation that analyses the skills to recommend a suitable job for the candidate. The solution
is to design a system that reads a resume and their skills. The resumes are going through pre-processing to make the design
more efficient. For pre-processing top words and porter stemmer, Porter Stemmer will make every word their root word, and
stop words will remove every meaningless word. This makes the system more efficient. Using tf-idf vectorizer for both resume
and job description. Then compare the skills in the resume and description.

For comparing, it uses the Cosine Similarity function and finds the scores of the resume for the respective jobs. Now it sorts
the list in descending order with respect to their scores. Now, he got a hierarchical order of jobs from top to bottom. So, he
can go with the first job or second which the skill he had already. He can be successful in that domain. The System not only
shows the job but also recommends the skills to be improved for the job. Because of this, the candidate can train
himself/herself for the future purpose and be a more achievable or talented person in his/her domain.

The proposed work focuses on predicting the suitable jobs for the candidates. It uses machine learning models to find
similarities between jobs description and resumes to predict accurately. This application can be used by any candidates who
need or who want to know about their suitable jobs and to improve themselves with both soft skills and hard skills. It will be
helpful to them by not wasting their time searching for jobs. They can also grow their skills in their domain and grow faster
in their domain.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a770

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
1.1 DATASET

Two Kaggle resources and some from google searches were combined to produce a Job Dataset. I gather resume data from my
friends. 13001 job descriptions are included in Job Dataset in .csv format. Each row has a different Id and three Features. The
features include Job Title, ID, Query, and Description. Another dataset exists, it is called the skill dataset. The information
for the skill dataset is gathered from Google by looking up each job. The job title and skills are included. There are 32 rows
of specific job skills listed.

Table 1.1 Summary of the dataset

Type of Dataset Number of rows

Job description 13000
Resumes 101
Job skills 32

1.2 METHODOLOGY

To find suitable jobs and their scores, this application receives the resume and has a dataset for a job with their description. It
will pre-process the resume and job description with the stop words and porter’s stemmer. Then it reduces into a meaningful
bag of words. Now the application uses a tf-idf vectorizer to convert a raw text into a matrix which makes it easy while
compare. The main step is comparing the two bag words. For that, it uses the Cosine Similarity function, which is an angle-
dependent calculation. By using cosine, it has a list of jobs in descending order with respect to scores. The system will move
on to the next progress which is finding the skills to be improved by the candidates. The system will take the resume and the
skills dataset then compares both and display the skills which are all not in the resume.

The major contribution of this work is as follows: The large MNC businesses use the mechanism currently in place for
employment recommendations. The method is employed by businesses, not by regular people. If not, they will charge a small
subscription fee to check the user's career options. The system functions for the average guy from city to village to modify this
predicament. Because the students would look for employment based on their own skills, this approach will reduce
unemployment. This company will also grow more quickly, which will result in more job openings.

The goal of the proposed work is to suggest a job that is ideal for the user. It displays the hierarchical jobs that are best for
the user, not just one job. Additionally, it suggests skills for the jobs that were suggested for the user. This project is intended
for someone who simply has no idea what they are going to do. Additionally, there are no logins available because doing so
increases the likelihood that users would reject you. The subsequent chapter goes over the specifics of the implementation.
The rest of the paper organizes as follows: Chapter 2 provides the literature review conducted for this project. Chapter 3
presents the System Design and Architecture of the project along with the methodology. Chapter 4 discusses the algorithms
proposed in this project. Chapter 5 presents the project conclusion and future works on this project.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a771

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
II. LITERATURE REVIEW

Existing works are mainly found for the company to select a candidate who is fit for their vacancy[17]. There are many
experiments for calculating the four recommendation algorithm but with a different distance formula namely the Minkowski
distance [5]. And some others are tried a different recommender system like collaborative which only helps when there are
more data to relate. That won’t help for a person who is searching that which job is the correct choice for him/her. R.J.
Mooney and L. Roy used Content-Based Book Recommending[1] where the content-based recommendation helps for a cold
start. And some authors also say that a content-based recommender is best when they researched a comparison study of job
recommendations [9].

A recommender system is not only the main part of accurate prediction. There are some other things like vectorizing the
words and then similarity functions. Authors like Shouning Qu[3], and Li-Ping Jing[2] said that for text mining, tf-idf is the
best approach for text feature selection. Ravali Boorugu has researched NLP and tried various text summarization
techniques[14][19]. Some papers also say about similarity detection with many languages 8][10]. Jeevamol Joy and Renumol
V G [16] discussed which similarity is the best one for a content-based recommending system. They finally concluded that
cosine similarity is the best similarity for content based recommended system. Cosine similarity is not only used for
recommender systems but is used to find the similarity functions between two sentences or two paragraphs[8][20].
Mohammad Alobed has tried “A Comparative Analysis of Euclidean, Jaccard, and Cosine Similarity Measure and
Arabic Wordnet for Automated Arabic Essay Scoring”[21], and L. Zahrotum also compared jaccarJaccardidean and cosine
similarity. They both said that Cosine similarity with all stemming types has the lowest error compared with the Jaccard and
Euclidean similarity[7]. There is already a system that worked with both tf-idf and cosine similarity recommendations. It is
used for patient support forums[6]. Tanya V. Yadalam, Vaishnavi, M. Gowda, and Vanditha researched those career
recommendations content-based filtering which was mostly like my project but inside it, they mostly discussed security,
transparency for the data, and the framework [15].

Most works are just built for the companies or for the purpose of making money from the people by giving some irrelevant
choices. Many were using collaborative recommendation, which recommends the many searched jobs or the jobs which
were chosen by some other. It only works if the system deals with more number of resumes which seems it can only be used
by the companies. Some systems are asking to log in and some were asked to buy subscriptions. Logging in makes you
redirect some spam mails.

In many papers, they have been solved through content recommender which is not enough. A literature paper[15] had done
research on content recommender system, tfidf vectorizer, and cosine similarity in a row but in that the author doesn’t think
about the implementation process and only concentrated more on securing the data.

III. SYSTEM DESIGN

In this chapter, it discusses about a quick overview of the system's architectural design and an explanation of the architecture.
Every module that is employed in the architecture design will be understood well.

3.1 OVERALL HIGH-LEVEL ARCHITECTURE

The high Lehigh-level architecture diagram Figure 3.1 displays the system's overall layout. A file with a job dataset contains
the name and description of the job. Resume data is information that the user entered.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a772

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882

Figure 3.1 High Level Architecture Diagram

The folder was read and then saved for later use. Using the tf-idf vectorizer, a matrix can be created from raw data for both
the employment dataset and the
user-collected resume. Then, using a method in similarity functions known as Cosine-Similarity, it will determine the scores
between the job description of the job dataset and the resume. It involves calculating the total of the vectors dot products
divided by the sum of their products divided by the products of their lengths. It will first display the top jobs and their scores
in table format before visualizing the results in a pie chart. The system will then analyze the user's resume and the skill dataset
to suggest skills that should be improved.

Figure 3.2: Modular View

The Modular view Figure 3.2 provides better visualization of the system design. Index is the area where the user uploads a
resume. Top Jobs will be recommended in descending order with scores. Visualizing the recommended jobs and scores.
Skills show the skills that need to be improved for the respective jobs.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a773

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
3.2 DETAILED DESIGN

3.2.1 Preprocessing

To obtain the cleaned dataset, it processes the data using Porter Stemmer and Stop words in Preprocessing Figure 3.3. "ID,"
"Query," "Job Title," and "Job Description" are all included in the Job dataset file. In this case, Query superset Job Title as
the job. Therefore, only the superset portion, Query, and its explanation are required in this system.

Figure 3.3: Preprocessing

To compare, the system doesn’t require terms like "is" or "are" that don't give a score or points, therefore it only reads or
chooses those two traits. Therefore, Stop words are employed to eliminate all the pointless/meaningless words. Porter
Stemmer then used to lower the word frequency. An algorithm for stemming is Porter Stemmer. Stemming is the process of
stripping a word down to its root, or lemma, which attaches to suffixes, prefixes, or other roots to form a base word. For
instance, the basic forms run, runs, running, and ran are all variations of the same basic form; run is the lemma. In the
appendices, the Porter Stemmer algorithm and comprehensive version are written. In this module, the old dataset will be
processed to produce a new, cleaned dataset that contains just lemma terms. Utilizes nltk (Natural Language Toolkit).
3.2.2 Similarity Function Module

A resume that was obtained from a user in text format and a cleaned employment dataset make up the Similarity Function
Module Figure 3.4. There is also a resume validation that, when the user tries to select a file, only displays the (.txt) file in
the file manager. The machine selected cosine similarity with tf-idf vectorizer for this process.

Figure 3.4 Similarity Function Module

The resume and the cleaned job dataset are both included in the Cosine Similarity function, which is processed by stopwords
and porterStemmer (which was also used to analyse the job dataset's description). Now, a processed text file is also a resume.
Use tf-idf, which stands for term frequency-inverse document frequency. Tf-idf was applied to both the processed resume
and description. We are given a matrix of values that is already included in the raw text. The more often a term appears in a
document, the higher it is regarded by the tf-idf. In order to account for the fact that a few terms appear more frequently
overall, it is balanced by the number of documents in the corpus that contain the word. After receiving a matrix value system,
the cosine similarity formula can be used to continue. The formula is the sum of the vectors' dot products divided by the sum
of the vector products divided by the vector products of lengths. Once the algorithm is operational, the system can produce
IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a774
www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
scores indicating how closely the job description and the resume match. Change the order of the scores that it is descending
to display the results. so that the user may quickly select the top-rated jobs and focus on them.

3.2.3 Skills Recommending Module

The Skills Recommending Module Figure 3.5 will demonstrate the workflow of the system that uses resumes and a skill
dataset to identify the top five skills for each position. The following paragraphs go into the implementation's details.

Figure 3.5: Skills Recommending Module

The skill dataset and resumes are the inputs for this module. Skills for the jobs are contained in a skill dataset. The algorithm
will now compare the word list for both the skill dataset and the resume. Additionally, print the top five talents that don't
match the resume. The system will list the top 5 occupations and top 5 abilities that the resume holder needs to develop in
this module. There are several cleaning procedures for the skill dataset, such as deleting the prefix and suffix spaces used to
match the words.
IV. PSEUDOCODE AND ALGORITHM

4.1 PREPROCESSING

The pre-processing module will clean the data of the job description and resume and give cleaned data. The
algorithm is described in Algorithm 4.1.

Input: Job dataset (.csv file)

Output: Cleaned Job dataset (.csv file)

Algorithm 4.1 Preprocessing

1: read job dataset file

2: reading a file with pandas and store it in a variable k
3: separate only job and description attributes from k by using split in python
4: By using split operation it turns into list 5: for each word
in description
6: if word not in stopwords then
7: use the word into porterstemmer
8: then convert the list of stemmed words into a txt format 9: else left
10: save in a separate file

4.2 SIMILARITY FUNCTION MODULE

The similarity function module will get the resume from user and scan it. Then compare the resume and cleaned job
description to give the top jobs and scores for it. The Algorithm for compressing the images and masks is described in
Algorithm 4.2.
IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a775
www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
Input: Cleaned job dataset and resume

Output: Recommending N jobs in descending order with respect to their scores.

Algorithm 4.2 Similarity Function Module

1: upload resume
2: if resume uploaded then
3: read and store a resume in a folder 4: else indicates to
enter a file
5: convert the resume into list and remove the spaces, tabs, and empty elements
6: then the pre-processing module to the resume file
7: apply both job description and resume matrices to cosine similarity formula.
8: convert matrix into array
9: Job description array as arr1 and resume array as arr2 10: for l1 in arr1 and
creating a list as a
11: for l2 in arr2
12: store sum of l1 in s1 and sum of l2 in s2
13: applying to formula s1.s2 / float(s1.s2)**0.5 and append to the list
14: Return
15: sum the duplicate jobs
16: store it in a dictionary then average it with the total number of count 17: sort in descending order
with respect to scores and display the results

The main work of my project were completed in Algorithm 4.2 and only recommending skills are left.
4.3 SKILLS RECOMMENDING MODULE
This module is used to recommend top 5 skills of top 5 jobs to the user to upgrade themselves. The Algorithm for
compressing the images and masks is described in Algorithm 4.3.

Input: Processed resume and skill dataset(.csv) file

Output: Top 5 skills for top 5 jobs

Algorithm 4.3 Skills Recommending Module

1: read the file skill dataset

2: reading a file with pandas and store it in a variable k 3: appending job to a
separate list and skills to a separate list 4: zip the job and skills from skill dataset
5: pick the top 5 jobs from the output of similarity function module and save it in list
6: for element in top5jobs
7: skills = values of element from the skill dataset 8: for element in
skills
9: if element not in listofResume 10:
store in recskills list
11: else ignore
12: slice the recskills list with only 5 elements 13: display the job
and skills to be improved

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a776

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
5 EXPERIMENTAL RESULTS AND OUTPUTS

The experiments and results of job a recommendation system with skills are discussed below.

5.1 PREPROCESSING

The job dataset will be processed by the system first. Using stop words and the porter stemmer process can be accelerated
and streamlined.

5.1.1 Job Dataset

There are four characteristics: ID, Query, Job Title, and Description. The system will reduce it to only two attributes, Query
and Description. Comparing and addressing job descriptions, after that, eliminate any words with no meaning or score using
stop words. The words are then transformed into stem words using porter’s stemmer prefixes and suffixes are eliminated.

5.2 SIMILARITY FUNCTION MEASURES

This module explains why the system chose to use Cosine similarity over other similarities. Cosine similarity, Euclidean
similarity, and Jaccard similarity were the three similarities that the system examined. Additionally, the same cleaned Job
dataset and a resume in (.txt) format were used to test these three similarities. The experiments involving the application of
the three similarities are discussed in the chapter that follows.

5.2.1 Cosine Similarity Function

The resume and the cleaned job dataset are both included in the Cosine Similarity function, which is processed by stopwords
and porterStemmer (which was also used to analyze the job dataset's description). Now, a processed text file is also a resume.

Figure 5.1: Cosine Similarity Function

The resume and the cleaned job dataset are both included in the Cosine Similarity function, which is processed by stop words
and porter Stemmer (which was also used to analyze the job dataset's description). Now, a processed text file is also a resume.
Use tf-idf, which stands for term frequency-inverse document frequency. Tf-idf was applied to both the processed resume
and description. It is given a matrix of values that is already included in the raw text. The more often a term appears in a
document, the higher it is regarded by the tf-idf. In order to account for the fact that a few terms appear more frequently
overall, it is balanced by the number of documents in the corpus that contain the word. After receiving a matrix value system,
the cosine similarity formula can be used to continue. The formula is the sum of the vectors' dot products divided by the sum
of the vector products divided by the vector products of lengths. Once the algorithm is operational, the system can produce
scores indicating how closely the job description and the resume match. Change the scores order such that it is descending
to display the results. so that the user may quickly select the top-rated jobs and focus on them. For eg. (the first job title is
data scientist and there are 400 data scientist’s description and it shows scores for every description). Then sum the scores of
a job which are similar and divide it with the total number of description (i.e 400). After that store the job and description in
dictionary has keys and values. Display the output by changing the scores into descending order. So that user can easily pick
the top rated jobs and concentrate to it.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a777

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882

Output :

Figure 5.2: Cosine Similarity in Array

The points in the array are shown in the previous picture Figure 5.2. This displays scores for all jobs, not just unique jobs.
The array value should then be averaged to determine a score for each individual jobs.

Figure 5.3: Cosine Similarity Output

5.2.2 Euclidean Similarity Function

The Euclidean Similarity also consists of the same input which was given
to the Cosine similarity function. The same process goes under Euclidean also but in cosine it used tf-idf vectorizer but here
it uses count vectorizer. It is used to transform a given text into a vector on the basis of the frequency (count) of each word
that occurs in the entire text. After converting the text into matrix value system is ready to proceed in a Euclidean similarity
formula.

Euclidean Similarity = 1 / (1+(d(x,y))) d(x,y) = √[ (x2 – x1)2 + (y2 – y1)2]

where, x and y are coordinates. (5.1)

As soon as the formula begins to produce results, scores that indicate how closely the job description and the resume match
are displayed. Change the scores' order such that it is descending to display the results. so that the user may quickly select
the top-rated jobs and focus on them.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a778

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
Output :

Figure 5.4: Euclidean Similarity Output

5.2.3 Jaccard Similarity Function

The identical input used by the Euclidean and Cosine similarity functions is also used by the Jaccard similarity function. The
only difference is in the metrics the cosine and Euclidean similarity data flows are identical. Here, the system makes no use
of a vectorizer; it merely measures the number of words in both the processed job description and processed resume. Apply
the Jaccard formula next. The number of observations in both sets divided by the number in either set is the Jaccard formula.
To put it another way, |AB| / |AB|.
As soon as the formula begins to produce results, scores that indicate how closely the job description and the resume match
are displayed. Change the scores' order such that it is descending to display the results. So, that the user may quickly select
the top-rated jobs and focus on them.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a779

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
Output :

Figure 5.5: Jaccard Similarity Output

5.2.4 Cosine or Euclidean or Jaccard

The positions that are advised for each of the three similarity functions. The system will now determine which is superior.
Offered my guide or mam a CV, the job list, and requested her to order the position. The top ten jobs out of thirty two were
ranked by mam, who then verified it with the three outputs. Mam's perspective and the cosine similarity recommendation
are practically identical. It compares the Cosine, Euclidean, and Jaccard similarity functions in the research paper.
Mohammad Alobed said in the paper that cosine similarity is superior to the other two. A Comparative Analysis of Euclidean,
Jaccard, and Cosine Similarity Measure and Arabic Wordnet for Automated Arabic Essay Scoring" is the title of the paper
that is related to the literature review.

Figure 5.6: Resume in List

The above figure 5.6 is showing the processed resume which is in list format.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a780

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882

Figure 5.7: Skills in List

The above figure 5.7 is showing the processed skills which are in list format.

Expert Recommended Output:

Table 5.1 Expert recommended top 10 jobs

S.NO JOBS
1. WEB DEVELOPER
2. FULL STACK DEVELOPER
3. SOFTWARE DEVELOPER
4. SOFTWARE ENGINEER
5. DATA VISUALIZATION EXPERT
6. SOFTWARE TESTER
7. DATA SCIENTIST
8. DATA ANALYST
9. INFORMATION SECURITY ANALYST
10. MACHINE LEARNING

The Cosine Similarity Function is the best to use, the system can conclude after comparison with the table. Therefore, cosine
was implemented, and the user interface showed the output. All of those will be seen in the following pages.

5.3 VISUALIZATION

After getting the top-recommended jobs and their scores. The system will display it as a pie chart. Google Charts were
utilized for visualization. Pie-chart has been constructed by using the keys and values from the Similarity Functions solution.
It is a three-dimensional pie chart. The label is included in a three-dimensional pie chart, and it is distinguished by different
colors. The job names will then be listed at the side of the three-dimensional pie chart in descending order according to the
result of cosine similarity functions. Moreover, a drop-down arrow will be present. The drop-down arrow is used to display
all thirty-two jobs that appeared in the list.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a781

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882

Figure 5.8: Pie Chart Output

5.4 SKILL RECOMMENDATION

A processed Resume and a skill dataset are required for skill recommendation. A resume with stop words and a porter
stemmer will be processed and turned into a list of words. Job and skills are the two attributes of the skill dataset.

Actual Input

Figure 5.9: Skill Data

Skill Data Figure 5.9 display the skills required for every job which are all included.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a782

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
Actual Output

Figure 5.10: Recommended Skills Output

VI. CONCLUSION AND FUTURE WORK

Job Recommendation System has a major role to play among recommending systems. With the presence of new
algorithms and techniques, the system needs to evolve along with it. The main objective of this project is to recommend
a suitable job for the candidates.

This project has two pre-processing methods, one text mining method and one similarity function. The pre-processing
methods are stop words and porter stemmer. The text mining method is tf-idf. The similarity function is a cosine similarity
function. Pre-processing methods are used with resumes and with jobs description, to make the system more efficient by
avoiding some garbage words. Tf-idif is used in processed resumes and processed jobs descriptions to convert it from
text to matrix to compare. Cosine Similarity will measure the similarity between the resume and each job description.
Finally, it will display the scores for the jobs in a sorted way. There is also a pie chart which is used to visualize the
percentage of the scores which is got by the candidate for the jobs. Then use a list compare method to compare the resume
and job skills to recommend the skills to be improved by the candidate.

FUTURE WORK

In this proposed work, there is only job recommendations and skill suggestion for IT jobs. It can be improved by
suggesting jobs and skills for the Non – IT jobs. In the future, some can find a better choice to find similarity than a
cosine similarity. It makes the recommendation more accurate.

REFERENCES

[1] R. J. Mooney and L. Roy, "Content-Based Book Recommending Using Learning for Text Categorization," in
In Proceedings of DL ’00: Proceedings of the Fifth ACM Conference on Digital Libraries, New York, NY, pp. 13-20,
2000

[2] Li-Ping Jing, Hou-Kuan Huang, Hong-Bo Shi, “Improved feature selection approach TFIDF in text mining”,
International Conference on Machine Learning and Cybernetics, pp. 944-946, 2002, doi:10.1109/icmlc.2002.1174522.

[3] Shouning Qu ,Sujuan Wang,Yan Zou, ” Improvement of Text Feature Selection Method Based on TFIDF”,
International Seminar on Future Information Technology and Management Engineering, pp. 79-81, 2008,
doi:10.1109/fitme.2008.25.

[4] I. A. Braga, "Evaluation of stopwords removal on the statistical approach for automatic term extraction,"
Seventh Brazilian Symposium in Information and Human Language Technology, pp. 142-149, 2009.

[5] Nikolaos D. Almalis, Prof. George A. Tsihrintzis, Nikolaos Karagiannis, Aggeliki D. Strati, “A new content-
IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a783
www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
based job recommendation algorithm for job seeking and recruiting”, 6th International Conference on Information,
Intelligence, Systems and Applications (IISA), pp. 1-7, 2015, doi:10.1109/iisa.2015.7388018.

[6] Mohammad Alodadi and Vandana P. Janeja, “ Similarity in Patient Support Forums Using TF-IDF and Cosine
Similarity Metrics”, International Conference on Healthcare Informatics, pp. 521-522, 2015, doi:10.1109/ichi.2015.99.

[7] L. Zahrotun, “Comparison jaccard similarity, cosine similarity and combined both of the data clustering with
shared nearest neighbor method,” Computer Engineering and Applications Journal. vol. 5. Pp. 11-18, 2016,
doi:10.18495/comengapp.v5i1.160, 2016.

[8] Peng Yi, Cheng Yang ,Chen Li, Yingya Zhang, “A Job Recommendation Method Optimized by Position
Descriptions and Resume Information”, IEEE Advanced Information Management, Communicates, Electronic and
Automation Control Conference (IMCEC), pp. 762 -764, March 2017, doi:10.1109/rteict.2017.8256590.
[9] Minh-Luan Tran, Anh-Tuyen Nguyen, Quoc-Dung Nguyen, Tin Huynh, “A comparison study for job
recommendation”, International Conference on Information and Communications (ICIC), pp. 199-204,
2017, doi:10.1109/infoc.2017.8001667.
[10] Gokul P.P, Akhil BK, Shiva Kumar K.M, “Sentence similarity detection in Malayalam language using cosine
similarity”, 2nd IEEE International Conference on Recent Trends in Electronics, Information & Communication
Technology (RTEICT), pp. 221-225, 2017, doi:10.1109/rteict.2017.8256590.

[11] Leah G. Rodriguez, Enrico P. Chavez, “Feature Selection for Job Matching Application using Profile
Matching Model”, IEEE 4th International Conference on Computer and Communication Systems (ICCCS), pp. 2-4 , Feb.
2019.

[12] Nunik Destria Arianti, Mohamad Irfan, Undang Syaripudin, Dina Mariana, Neny Rosmawarni, Dian
Sa’adillah Maylawati, “Porter Stemmer and Cosine Similarity for Automated Essay Assessment”, 5th International
Conference on Computing Engineering and Design (ICCED)”, pp. 1-5, 2019,
doi:10.1109/icced46541.2019.91610.

[13] Garima Koushik, Dr. Prof. K. Rajeswari, Mr. Suresh Kannan Muthusamy, “Automated Hate Speech
Detection on Twitter. 5th International Conference On Computing, Communication, Control And Automation
(ICCUBEA)”, pp. 421- 425, 2019, doi:10.1109/iccubea47591.2019.912.

[14] Ravali Boorugu, Dr. G. Ramesh, “A Survey on NLP based Text Summarization for Summarizing Product
Reviews”, Second International Conference on Inventive Research in Computing Applications (ICIRCA), pp. 352- 356,
2020, doi: 10.1109/ICIRCA48905.2020.9183355.

[15] Tanya V. Yadalam,Vaishnavi, M.Gowda, Vanditha Shiva Kumar, Disha Girish, Namratha M, “Career
Recommendation System Using Content Based Filtering”, International Conference on Communication and Electronics
Systems (ICCES), pp. 2-5, June 2020, doi: 10.1109/ICCES48766.2020.9137992

[16] Jeevamol Joy and Renumol V G, “Comparison of Generic Similarity Measures in E-learning Content
Recommender System in Cold-Start Condition”, IEEE Bombay Section Signature Conference (IBSSC)”, pp. 175-
179, 2020, doi:10.1109/ibssc51096.2020.9332162.

[17] M. Alamelu, D.Sathish Kumar, R. Sanjana, J.Subha Sree, A.Sangeerani Devi,

D. Kavitha, “Resume Validation and Filtration using Natural Language Processing”, 10th International Conference on
Internet of Everything, Microwave Engineering, Communication and Networks (IEMECON), pp. 412-430, 2021,
doi:10.1109/IEMECON53809.2021.9689075.

[18] Swaranjali Jugran, Ashish Kumar, Bhupendra Singh Tyagi, Mr. Vivek Anand,”Extractive Automatic Text
Summarization using SpaCy in Python & NLP”, International Conference on Advance Computing and Innovative
Technologies in Engineering (ICACITE), pp. 582-585, 2021, doi:10.1109/icacite51222.2021.9404712.
[19] S. Prathyusha, S. Jadhav, K. Kommu, M.S. Velpuru, “Text summarization using NLTK with GUI interface”,
4th Smart Cities Symposium (SCS 2021), pp. 435-442, 2021, doi: 10.1049/icp.2022.0369.

[20] Meenakshi A. Thalor, “A Descriptive Answer Evaluation System Using Cosine Similarity Technique”,
International Conference on Communication information and Computing Technology (ICCICT), pp. 1-12, 2021, doi:
10.1109/ICCICT50803.2021.9510170.

[21] Mohammad Alobed;Abdallah M M Altrad;Zainab Binti Abu Bakar, “ A Comparative Analysis of Euclidean,

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a784

www.ijcrt.org © 2022 IJCRT | Volume 10, Issue 8 August 2022 | ISSN: 2320-2882
Jaccard and Cosine Similarity Measure and Arabic Wordnet for Automated Arabic Essay Scoring”, Fifth International
Conference on Information Retrieval and Knowledge Management (CAMP), pp.70-74, 2021,
doi:10.1109/camp51653.2021.9498119.

IJCRT2208099 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org a785