International Journal of Enhanced Research in Management & Computer Applications

ISSN: 2319-7471, Vol. 13, Issue 4, April-2024, Impact Factor: 8.285

Presented at “ICRETETM-2024”, Organized by GSMCOE, Pune, on 22nd - 23rd April 2024

Ai Resume Analyzer Using Natural Language

Processing and Data Mining
Jayshri Mankar*1, Abhishek Chauhan*2, Aniket Gophane*3, Aditya Karle*4,Taslimarif
Makandar*5, Akshay Funde*6
Department of Computer Engineering, G. S. Moze College of Engineering, Balewadi, Pune-411045, India

ABSTRACT

Imbalance data conversion into structured data is the very tedious task in data mining techniques, various techniques have
been already introduced to extract the data from large textand extract the features using various feature extraction techniques,
some machine learning algorithms have been already introduced by various researchers for classification and display the
results on heterogeneous data. This work suggests a method of eliminating or resuming important information in a
curriculum vitae from the semi-structured textformat, and rating it according to the preference and requirements of the client.
The whole process was divided into three basic sections to achieve the desired goal. The first section consists of segmenting
the whole Summary according to the content of each part, the second section consists of extracting data in a standardized
form from unstructured data and the final section consists of analyzing structured data using NLP and Machine learning
algorithms. The Stanford NLP rule extraction algorithm ha used to extract the various rules from raw data and select some
important feature for classification as well as optimization. Experimental analysis shows the effectiveness of proposed
system with classification accuracy.

Keywords: Resume parsing, Data Mining, Machine learning, NLP

INTRODUCTION

he need of thinking and designing an appropriate resume. Classification is specially essential in solutions to data processing
and machine-learning. Nowadays, many outlets have generated the numerous types of data in row format, as well as its
hard to process from existing environments and algorithms. Text classification requires assigning the text to one or more
predefined groups using some kind of classification algorithm performed by the content of the document. A Generic
classification corpus has been developed and a single assessment system has been introduced to identifyEnglish text based on
machine learning, which has now made significant progress. Most of the evidence in the real world is contained in relational
bases. Data clustering is an essential machine learning process in which a sub-set of candidate labels is allocated to an
entity, the main issue with multi-label clustering is the redundant online clustering method and the offline data set for
dealing with this issue. We plan to use unstructured data classification to structured conversion systems and maximize the
accuracy of the final sub-cluster. Demonstrate two implementations of our method using logistic regressions and improved
gradient trees, along with a simple procedure for Expectation Maximization preparation. We also get an efficient prediction
approach dependent on dynamics programming.
Resume Analyzer is software developed to simplify the task ofcreating a resume in structured format. The system is flexible
to be used and reduces t

LITERATURE REVIEW

According to [1] a recruiting case study as a basis for a statistical evaluation of several methods for calculating similarity
scores. To this end, we suggest using a computer- aided resume evaluator on a group of resumes, then has professionals
evaluate the same set of resumes, and finally look for a connection between the two sets of results. Finding the right
computer-aided resume evaluator for digital human resources requires a consideration of the various similarity score
calculation methodologies now available for processingresumes.

Page | 370
 International Journal of Enhanced Research in Management & Computer Applications
ISSN: 2319-7471, Vol. 13, Issue 4, April-2024, Impact Factor: 8.285
Presented at “ICRETETM-2024”, Organized by GSMCOE, Pune, on 22nd - 23rd April 2024
According to [2] an approach to resume writing that is both straightforward and straightforward to apply. We propose a
program that, given some basic information about the applicant, may generate a professional-looking résumé. Users may
sign up for an account and start working on their resume by entering their login details and receiving a one-time password
(OTP).

According to [3] Recruiters have a hard time finding the greatest fit for a job position since the resumes candidates submit
vary in format (e.g., font, color, font size, etc.). To combat these issues, recruiters may turn to natural language processing
(NLP) to glean the specifics about potential candidates they need to move their candidacy ahead. In this paper, we suggest
using the Stanford CoreNLP system's named entity recognition capabilities to glean data useful in the hiring process.

According to [4] Data-driven HR has been shown to significantly enhance the quality and speed of the whole recruitment
process by using Natural Language Processing tools. First, a resume parser has been built utilizing natural language
processing to evaluate the most important aspects of the hiring process. The computational framework of the parser was then
used to create a potent instrument for resume matching based on job requirements, with the candidate's ability to produce a
pie chart serving as an input.

According to [5] It might be challenging to find qualified candidates for an open position, particularly if there are numerous
applications to choose from. It may be difficult for the team to find the most qualified candidate at the most opportune time
if they have to go through each resume manually. An automated method of screening and rating applications might
significantly reduce the time spent on the screening process. The KNN algorithm is used to select and rank Curriculum
Vitaes (CV) based on job descriptions in large numbers, and the cosine similarity is used to find the CVs that are most
relevant to the provided job description in our work.

According to [6], They have propose a substance extraction approach for getting content from news pages that joins a
division like methodology and a thickness based methodology. A tool Block Extractor is used to identifies contents in three
steps. First, it looks for all Block-Level Elements and Inline Elements blocks, which are designed to roughly segment pages
into blocks. Second, it computes the densities of each BLE and IE block and its element to eliminate noises. Third, it
removes all redundant BLE and IE blocks that have emerged in other pages from the same site. Compared with three other
density-based approaches, our approach shows significant advantages in both precision and recall. BLE and IE blocks to
gather related noises or contents. Next, we used this density- based approach and redundancy removal to obtain the final
content. Based on our approach, a tool called Block Extractor was developed.

In this paper [7], The issue of naturally removing web information records that contain user generated content (UGC).To
solve this problem MiBAT and MDR algorithms are used 1)MiBAT (Mining data records Based on Anchor Trees). They
have represented two space imperative guided likeness measures, for example PM and PS. They have propose an
information record mining calculation utilizing either PM or PS. Our instinct is exceptionally basic: each record comprises
of one or a few sub trees, just one of which contains the rotate. We call such sub-trees that contain turns as grapple trees,
since they give stay point data about where information records are found. 2)MDR (Mining Data Records in Web pages)
MDR identifies a list of records by conducting a similarity test against a pre-defined threshold for two sub- trees in the
DOM tree of a web page. Such a method is referred to as the similarity-based approach, because the underlying assumption
is that data records belonging to the same list usually have similar DOM tree structures MDR and its Limitations:-A group
of similar objects, which forms a data region, is usually presented in a contiguous region and format- ted using similar
HTML tags. Every record in a data region is formed by the same number of adjacent child sub-trees under the same parent
node. Novel mining algorithm called MiBAT which makes use of domain constraints to acquire anchor pointinformation. Our
methodology accomplishes an exactness of 98.9% and a review of 97.3% concerning post record extraction. On page level,
it lawlessly handles 91.7% of pageswithout removing any o_-base posts or missing any brilliant posts.

This paper [8] depicts a framework for robotized continue data extraction to help fast resume search and the board. The
framework is equipped for extricating a few significant educational fields from a free arrangement resume utilizing a lot of
common language handling (NLP) strategies. We depict a working framework, for programmed continue the board. The
framework is equipped for extricating six significant fields of data as characterized by HR-XML In this the main layer is
made out of a few general data squares, for example, individual data, instruction and so forth. The second layer of structure
is inside the principal layer and contains explicit data comparing to the layer 1. For instance, the layer 1 individual data
square comprises of layer 2 data like name, address and email. While this probably won't be valid for every one of the
resumes, the structure is by all accounts held in the greater partof resumes. Furthermore, the area of the data (like name, age

Page | 371
 International Journal of Enhanced Research in Management & Computer Applications
ISSN: 2319-7471, Vol. 13, Issue 4, April-2024, Impact Factor: 8.285
Presented at “ICRETETM-2024”, Organized by GSMCOE, Pune, on 22nd - 23rd April 2024
and so forth) in resumes differs fundamentally from resume to continue. Our framework can chip away at both layered
structure and unstructured resumes. Data extraction module is made out of a few sub modules, every one of which plays out
the undertaking of removing explicit data. The primary sub modules are (a) Qualification module, (b) Skill module (c)
Experience module and (d) individual data extraction. While the capability extraction sub-module separates the graduating
college name, degree and the class acquired. The aptitudes extraction module extricates the abilities of the applicant.
Experience extraction module is competent or removing the all-out understanding, in any event, when this data isn't
expressly referenced in the resume of the candidate. The extraction procedure utilizes a lot of language preparing systems
which are part heuristics and part example coordinating. Test results completed on countless resumes demonstrate that the
proposed framework can deal with an enormous assortment of resumes in various record positions with an exactness of
91% and a review of 88%.

In this paper [9] we present a near investigation of 5 estimates utilizing distinctive vector loads done over an enormous
arrangement of French list of qualifications. The point is to know how these measures act and whether they approve the
idea that chose list of qualifications have more in a similar manner as themselves than with the dismissed list of
qualifications. We utilize NLP systems and ANOVAs to do the relative examination. The outcomes demonstrate that the
determination of measures and vector loads must not be viewed as insignificant in e-Recruitment projects; especially in
those where the resumes' resemblance is estimated. Something else, the outcomes may not be dependable or with the
normal execution. Four sorts of archives are dissected in this work: pdf, Microsoft Word, Open Document Text and Rich
Format Text.

In this paper [10] the general target of this examination was to concentrate such information as experience, highlights, and
business and training data from resumes put away in HR archives. In this article, we propose a philosophy driven data
extraction framework that is intended to work on a few million free-design printed resumes to change over them to an
organized and semantically improved rendition for use in semantic information mining of information basic in HR forms.
The engineering and working instrument of the framework, similitude of the idea and coordinating strategies, and a
deduction system are presented, and a contextual investigation is displayed.

RESEARCH METHODOLOGY

In our proposed study, a machine learning strategy based on a recommendation algorithm is used to predict student
achievement using both synthetic datasets and real-time student data. Consider figure for a thorough explanation of the
machine learning classifier-based proposed design. The proposed system architecture with the machine learning classifier is
shown in Figure 1. We first gathered information from a variety of sources, including different web applications, actual
student information, company information and a few synthetic data sets from diverse sources.

Fig 1: System Architecture of Proposed SystemList of Modules and Functionality

Page | 372
 International Journal of Enhanced Research in Management & Computer Applications
ISSN: 2319-7471, Vol. 13, Issue 4, April-2024, Impact Factor: 8.285
Presented at “ICRETETM-2024”, Organized by GSMCOE, Pune, on 22nd - 23rd April 2024

The proposed system follows below procedure for entire execution to convert unstructured data to structured conversion in
entire process.

 Initially some raw resume data has given input to system, it should be unstructured format. (it should be doc, pdf file)
 Read data from entire document and apply stop word removal as well as porter stemming algorithm to get the lemmas
features.
 Natural Language Processing (NLP) is another technique has used to extract the features from text using dependency
parser.
 To identify the name of specific entity has used NameEntity Recognizer of Stanford NLP.
 Once semi-structured format has done, selected feature has place in structured format and using any respective
machine learning algorithm.
 Once clarification has done we calculate the confusion matrix for entire test data and predict the precision, recall,
accuracy etc. respectively.

ALGORITHM DETAILS:

Similarity (Machine learning) Algorithm:

Input: Training set with established standards or guidelines Performing a test using data that has been normalised from the
Train_Data. Test_Data indicated the threshold Th_values.

Output: Output set using the parameters {Predicted_class, weight_Score}

Step 1: The following code can be utilised to read each test result from the Test_Data array and verify training criteria.
Subsequently, the data is standardised and modified to align with the requirements of the algorithms.

Step 2: Select the characteristics from the extracted attributes set of the test. Using the provided code, generate a feature
map by utilising the data as features.

Test_Feature_List [t…..…n] = Σ(t)nx=1 test_Feature(x)

Test_Feature_List [x] contains the features that were chosen for further inspection.

Step 3: Subsequently, it is necessary to thoroughly examine the entire training dataset to design the rule employed to
categorise all test data.

Step 4: Generate the training dataset by utilising the provided function on the input dataset.

Train_Feature_list [t…..…n] = Σ(t)nx=1 train_Feature(x)

Page | 373
 International Journal of Enhanced Research in Management & Computer Applications
ISSN: 2319-7471, Vol. 13, Issue 4, April-2024, Impact Factor: 8.285
Presented at “ICRETETM-2024”, Organized by GSMCOE, Pune, on 22nd - 23rd April 2024
The enumerated regulations The Train_Feature_list[t] generate the feature vector that is utilised to build the hidden layer.
This utilizes the train data to evaluate all test cases.

Step 5: The similarity weight is determined once the feature map has been generated.

Step 6: Assess the disparity between your current weight and your desired weight.

if(Gen_weight> =qTh)

Step 7: Out_List.add (trainF.class,weight)

Step 8: Advance to step 1 and proceed once the Test is finished. The data variable has a null value.

Step 9: ReturnOut_List.

1 : Stop word Removal Approach

Input: Stop words list L[], String Data D for remove the stop words.
Output: Verified data D with removal all stop words.
Step 1: Initialize the data string S[].
Step 2: initialize a=0,k=0
Step 3: for each(read a to L)

If(a.equals(L[i]))
Then Remove S[k]
End for
Step 4: add S to D.
Step 5: End Procedure

2 Stemming Algorithm.
Input : Word w
Output : w with removing past participles as well.
Step 1: Initialize w
Step 2: Intialize all steps of Porter stemmer
Step 3: for each (Char ch from w)
If(ch.count==w.length()) && (ch.equals(e))
Remove ch from(w)
Step 4: if(ch.endswith(ed))
Remove ‘ed’ from(w)
Step 5: k=w.length()
If(k (char) to k-3 .equals(tion))
Replace w with te.
Step 6: end procedure.

RESULTS

The partial implementation of proposed system has been completed for the training module. As per our first module wehave
used standard pdf data set of 2 student data files for dataset. The below figure 2 shows the ranking show of whole data

Page | 374
 International Journal of Enhanced Research in Management & Computer Applications
ISSN: 2319-7471, Vol. 13, Issue 4, April-2024, Impact Factor: 8.285
Presented at “ICRETETM-2024”, Organized by GSMCOE, Pune, on 22nd - 23rd April 2024

Fig 2: Ranking Graph

As per our module we have used standard Input pdf data set . The below figure 4 shows the output file of whole data

Fig 3: System Input File Format

Page | 375
 International Journal of Enhanced Research in Management & Computer Applications
ISSN: 2319-7471, Vol. 13, Issue 4, April-2024, Impact Factor: 8.285
Presented at “ICRETETM-2024”, Organized by GSMCOE, Pune, on 22nd - 23rd April 2024

Fig 4: System Output File Format

CONCLUSION

Based on the proposed experimental analysis this system will provide better and efficient solution to current hiring process.
This will provide potential candidate to the organization and the candidate will successfully be placed in an organization
which appreciates users skill set and ability and speed up the whole hiring process. To work with various kinds of large
unstructured data will be future work for such systems.

REFERENCES

[1]. Özçevik, Yusuf, Fatih Yücalar, and Murat Demircioğlu. "Determining a Proper Text Similarity Approach for
Resume Parsing Process in a Digitized HR Software." Celal Bayar University Journal of Science 18.4 (2022): 371-
378.
[2]. Tyagi, Rinki, et al. "Resume Builder Application." International Journal for Research in Applied Science and
Engineering Technology (IJRASET) Volume 8 (2020).
[3]. Mittal, Vrinda, et al. "Methodology for resume parsing and job domain prediction." Journal of Statistics and
ManagementSystems 23.7 (2020): 1265-1274.
[4]. Deepak, Gerard, Varun Teja, and A. Santhanavijayan. "A novel firefly driven scheme for resume parsing and
matching based on entity linking paradigm." Journal of Discrete Mathematical Sciences and Cryptography 23.1
(2020): 157- 165.
[5]. Tejaswini, K., et al. "Design and development of machine learning based resume ranking system." Global
Transitions Proceedings 3.2 (2022): 371-375.
[6]. Shuang Lin, Jie Chen, Zhendong Niu, \Combining a Segmentation-Like Approach and a Density-Based Approach
in Content Extraction \ , August 2017.
[7]. Xinying Song, Jing Liu, Yunbo Cao, Chin-Yew Lin, and Hsiao-Wuen Hon , "Automatic Extraction of Web Data
Records Containing User-Generated Content", Jan. 2015.
[8]. Sunil Kumar Kopparapu , \Automatic Extraction of UsableInformation from Unstructured Resumes to Aid Search",
Page | 376
 International Journal of Enhanced Research in Management & Computer Applications
ISSN: 2319-7471, Vol. 13, Issue 4, April-2024, Impact Factor: 8.285
Presented at “ICRETETM-2024”, Organized by GSMCOE, Pune, on 22nd - 23rd April 2024
Oct.2016.
[9]. Paul, P. Efficient Graph-Based Document Similarity. In: The Semantic Web. Latest Advances and New Domains,
Springer International Publishing, 2016, pp 334–349.
[10]. Wang, J, Dong, Y. 2020. Measurement of Text Similarity: A Survey. Information, vol 11(9).
[11]. Farouk, M. 2019. Measuring Sentences Similarity: A Survey. Indian Journal of Science and Technology, 12(25),
1–11.
[12]. Yujian, L, Bo, L. 2007. A Normalized Levenshtein Distance Metric. IEEE Transactions on Pattern Analysis and
Machine Intelligence, vol 29 (6), pp 1091-1095.
[13]. Dreßler, K, Ngomo, A, N. 2017. On the Efficient Execution of Bounded Jaro-Winkler Distances. Semantic Web,
vol 8, pp 185-196.
[14]. Del, M, Angeles, M, García-Ugalde, F, Valencia, R, Nava, A. Analysis of String Comparison Methods During
DeDuplication Process. International Conference on Advances in Databases, Knowledge, and Data Applications,
Rome, Italy, 2015, pp 57-62.
[15]. Bakkelund, D. 2009. An LCS-based string metric. Olso, Norway: University of Oslo. [19]. Kondrak, G. N-Gram
Similarity and Distance. In String Processing and Information Retrieval, Springer Berlin Heidelberg, 2005, pp
115–126.
[16]. Luis Adri_an Cabrera-Diego1,2, Barth_el_emy Durette2, Matthieu Lafon2, Juan-Manuel Torres-Moreno1,3 and
Marc El-B_eze1, "How Can We Measure the Similarity Between R_esum_es of Selected Candidates for a Job?
Luis Adri_an Cabrera-Diego1,2, Barth_el_emy Durette2, Matthieu Lafon2, Juan-Manuel Torres- Moreno1,3 and
Marc El-B_eze1" , Jan.2014.
[17]. Duygu C_EL_IK,"Towards a semantic-based information extraction system for matching resumes to job openings",
June2016.

Page | 377