Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)

IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

An Advanced Real-Time Job Recommendation

System and Resume Analyser
B. Leela Prasad 1 K. Srividya 2 K. Nikhil kumar 3
Dept.Of CSE Dept.Of AIML Dept.Of CSE
GMR Institute Of Technology GMR Institute Of Technology GMR Institute Of Technology
Rajam, India Rajam, India Rajam, India
[email protected] [email protected] [email protected]
2023 International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS) | 979-8-3503-0085-7/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICSSAS57918.2023.10331788

L. Kiran Chandra 4 N. S.S.K Dil 5 G. Vamsi Krishna 6

Dept.Of CSE Dept.Of CSE Dept.Of CSE
GMR Institute Of Technology GMR Institute Of Technology GMR Institute Of Technology
Rajam, India Rajam, India Rajam, India
[email protected] [email protected] [email protected]

Abstract— The increasing complexity and abundance of current contextual embedding methods are not comfortable
information in resumes have created confusion and challenges for dealing with large size resumes. This paper suggests an
for both candidates and recruiters. An effective resume approach using domain adaptation technique based on a graph
screening is crucial for students looking for job opportunities as neural network to categorise resumes without needing labelled
they can demonstrate their skills and qualifications to potential
employers and is essential for the recruiter. This study focuses
resumes and retraining the model to handle these issues [1].
on implementing a resume screening and job recommendation The enrolment process for organizations has changed due to
system using Graph Neural Network based on domain the internet, with online job listings attracting a wide range of
adaptation approach and Natural language processing. The goal candidates. However, this has also brought about difficulties
is to extract latent features from job posts and resumes using like the submission of several irregular and diverse resumes
GNN and NLP techniques, including Named Entity Recognition [5]. Utilizing machine learning algorithms and web scraping
(NER). Furthermore, the system leverages the GloVe word techniques to streamline various tasks, including skill
embeddings for tokenizing and encoding textual information, assessment, resume screening, job notifications, and interview
enhancing semantic understanding. In addition, the system arrangement. This system encompasses functionalities like
evaluates the resume score based on the job description using
advanced techniques such as cosine similarity and TF-IDF to
conducting virtual practice interviews, automated
assess the relevance of keywords and phrases between the appointment scheduling, and administering a pre-interview
resume and job description. By considering the job roles questionnaire, all while monitoring applicants' backgrounds
matched to the resume it also recommends the latest jobs posted [4].Traditional text representation methods suffer from sparse
on different websites using web scraping. In conclusion, this and high-dimensional representations, while deep learning
study demonstrates the effectiveness of the GNN-based domain models like CNN and RNN have limitations in capturing
adaptation method can make more accurate in resume global word co-occurrences. Graph Neural Networks (GNNs)
screening and job recommendations. have emerged as a promising approach for text classification,
as they can capture global information through message
Keywords— NLP, GNN, Domain Adaption, Job passing over graphs [6]. The advancement of communication
recommendation and Resume Analyzer. and information technology has paved the way for the internet
to become a prominent tool for securing employment during
college campus recruitment periods. A novel hybrid job
I. INTRODUCTION recommendation approach is employed within college
recruitment systems, integrating collaborative filtering and
Resume screening and Job recommendation systems user profile-based filtering. This approach exhibits greater
plays a vital for both candidates and recruiters. Due to accuracy when compared to conventional recommendation
increasing in technologies a large number of jobs are created systems [11][12].
and new trends comes into market every day. For a candidate
to apply for a job, it is crucial to understand which work roles LLMs' comprehensive contextual data and semantic
are compatible. And also, it is difficult to evaluate the resume representations allow for the analysis of behaviour graphs and
and job description manually whether the resume is matched the discovery of patterns and linkages. It introduces a meta
for particular job role or not. So, to make the process better path prompt constructor to encode the interaction information
and quicker an AI based resume analyzer and Job of the graph into natural language prompts. The framework,
called GLRec (Graph-understanding LLM Recommender), is
recommendation system is needed for the candidate.
fine-tuned with LoRa in a constructed instruction dataset to
optimize the quality of job recommendations. Using Large
Many of the traditional algorithms proved their Language Model can enhance the screening process and
efficiency in resume classification, but they require labelled recommendation systems [8]. With the help of Large
data, which is scarce and difficult to obtain. Additionally, the

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 1039

Authorized licensed use limited to: Andhra University College of Engineering. Downloaded on February 18,2025 at 05:17:32 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

Language Models (LLMs) and Generative job G. M., & Suganthi, S. K. [7] developed AI based system that
Recommendation based on Large language models (GIRL) predict a suitable candidate. Additionally, four clusters
can creates personalized job descriptions for job seekers based derived from the job description were utilised to measure
on their Curriculum Vitae (CV). The generated job similarity and correspond to primary skills. Three classifiers,
descriptions serve as references for job seekers. They also such as Adaboost, linear regression, decision trees, and
bridge the semantic gap between CVs and job descriptions, XGBoost are used to determine candidate suitability. The
improving the performance of traditional discriminative Jaccard similarity measure is used to evaluate the suitability
models [9]. between a Job Description (JD) and a Candidate Resume
II. RELATED WORK (CR).

Wu, L., Qiu, Z., Zheng, Z., Zhu, H., & Chen, E. [8]
Chung, Y. C., & Kuo R J. [1] introduced an domain have introduced Large Language Models (LLMs) as a way to
adaptation approach, utilizing a graph neural network comprehend behavioural graphs and apply this
framework for the automated classification of resumes within comprehension to enhance recommendations in the context of
database systems. The primary goal is to uncover underlying online recruitment. They have also presented a novel
characteristics of job postings, enabling the categorization of framework known as GLRec, which integrates Large
resumes without the need for labelled resume data. The Language Models (LLMs) with behavioural graph
outcomes revealed that the suggested method had a high comprehension to elevate the quality of job recommendations.
accuracy of 81%. But one of the major limitations observed in Their assessment of the framework's performance utilizes the
this proposed method is difficult to enhance the model's AUC metric.. Zheng, Z., Qiu, Z., Hu, X., Wu, L., Zhu, H., &
performance, such as studying the autocorrelation of Xiong, H. [9] proposed a novel approach called Generative
occupational groups and classifying resumes into proper job Recommendation based on Large language models for job
categories. Spoorthi, M., Kuppala, M., Karpe, V. S., & recommendation systems with Large Language Models
Dharavath, D. [2] presents an automated resume classification (LLMs) to generate personalized job descriptions for job
system that incorporates deep learning models for the seekers. And evaluates the performance of the GIRL approach
categorization of resumes, focusing on the skills mentioned using two different metrics: AUC and Log Loss. Du, Y., Luo,
within each resume. The system employs an ensemble D., Yan, R., Liu, H., Song, Y., Zhu, H., & Zhang, J. [15]
learning approach, where the accuracy of the system is devised an interactive approach for completing resumes,
contingent upon the quality of the training data. It's worth addressing the constraints associated with the synthetic
noting that the system's effectiveness could be compromised generation of Large Language Models (LLMs) and the use of
if the training data lacks accuracy or carries biases. Gu, Y., Generative Adversarial Networks (GANs) in refining
Wang, Y., Zhang, H. R., Wu, J., & Gu, X. [3] suggest an representations of subpar resumes. This method incorporates
innovative approach to text classification called Text-MGNN, an alignment multi-objective learning framework tailored for
which leverages Graph Neural Networks (GNNs) to classify job recommendation.
text by considering multiple levels of granularity while also
incorporating a topic-aware perspective. It is evaluated on Li, S., Li, K., & Lu, H. [10]examine the potential bias
three real-world datasets and outperforms 11 baseline in deep-learning-powered automated resume screening tools,
methods. The proposed method archives 97% accuracy with specifically in the use of word embedding for low-
R8 dataset and with R52 it archives 94% accuracy. Silva, G. dimensional vectors. The use of different values of the
L. L., Jayasinghe, T. L., Rangalla, R. H. M., Gunarathna, W. boundary parameter (τ) and the regularization parameter (λ) to
K. L., & Tissera, W.[4] developed an system that helps to evaluate the performance of the bias mitigation method. Zhu,
identify the skills and generate resumes. The objective is to Y. [11] proposed a hybrid approach for job recommendation
use machine learning algorithms and web scraping technology algorithm for college employment systems. Shaikym, A.,
to automate processes like skill matching, shortlisting of Zhalgassova, Z., & Sadyk, U. [12] design a job
resumes, and scheduling interviews. recommendation system and evaluated using a hybrid
approach. The system uses advanced techniques and
Ransing, R., Mohan, A., Emberi, N. B., & algorithms. Sharma, M., Choudhary, G., & Susan, S. [13]
Mahavarkar, K. [5] employ machine learning techniques, introduced a method for efficient classification of resumes.
including K-Nearest Neighbour (KNN), Linear Support Also introduce Elite bag-of-words, a text vectorization
Vector Classifier (Linear SVC), and XGBoost, to construct a technique for the vectorization of resumes and compare its
ranking system aimed at suggesting more suitable resumes performance with other existing bag-of-words approaches.
based on textual job descriptions. They introduce a two-tiered Petersheim, C., Lahey, J., Cherian, J., Pina, A., Alexander, G.,
stacked model of base models, which yields promising & Hammond, T. [14] introduced a comparative evaluation of
outcomes when contrasted with using individual models and resumes from various domains. Their findings indicated that
got an accuracy of 85 %. Furthermore, the authors note the students were 7.4% more inclined than recruiters to advance
potential for performance improvement by incorporating deep resumes to the next stage. Additionally, students spent an
learning models such as Long Short-Term Memory (LSTM), average of 7.2 seconds less time reviewing each resume.
Recurrent Neural Networks (RNN), or Convolutional Neural Furthermore, they observed that for each incremental point
Networks (CNN). Tripathi, R. C[6] proposed method increase in GPA, there was a 26.6-28.5% higher likelihood of
designed to streamline the labour-intensive and equitable a resume being advanced.
screening process through intelligent automated resume
analysis and recommendations and got 87% accuracy. Sridevi,

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 1040

Authorized licensed use limited to: Andhra University College of Engineering. Downloaded on February 18,2025 at 05:17:32 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

Wosiak, A. [16] create a system specifically resumes. On the other hand, the second dataset is Jobs on
designed for Polish resume documents, with a primary focus Naukri.com, has a broader scope, aimed at developing a Job
on optimizing the IT recruitment process. The system's goal is and Skills Recommendation System. This dataset is
to enhance candidate screening and selection by improving considerably larger, comprising 27,010 records and
efficiency and accuracy. Pimpalkar, A., Lalwani, A., encompassing seven fields like Job Salary, Job Experience
Chaudhari, R., Inshall, M., Dalwani, M., & Saluja, T. [17] Required, Key Skills, Role Category, Functional Area,
implement NLP techniques to enable resume selection and Industry, and Job Title Category in addition to the Resumes
identification, with a focus on analysing resumes. It consists field. This dataset is leveraged to offer job recommendations
of machine learning algorithms like SVM and Random forest and skill matching based on job listings, making it a valuable
that gives accurate results of 82% in resume classification. resource for enhancing the recruitment and job search
Pant, D., Pokhrel, D., & Poudyal, P. [18] developed an experience.
automated screening system tailored for software engineering
position resumes. This system harnesses natural language
B. Data Preprocessing
processing, word matching, character positioning, and regular
expression techniques. During a random test involving ten Data preprocessing plays a pivotal role in getting raw
distinct resumes, the system successfully extracted and text data ready for analysis in the context of the provided
summarized 70.59% of the skills mentioned in those resumes. datasets. This multifaceted process involves systematically
Venkataramana, A., Srividya, K., & Cristin, R. [19] introduces applying various techniques to prepare raw text data for
BART, a Deep Learning model, to address this challenge. analysis, model training, and information extraction. It starts
BART merges encoder and decoder components and employs with text cleaning, which involves removing special
an attention mechanism to highlight key data features. It's characters and punctuation, ensuring that the text is free from
compared with other models, including BERT, T5, and elements that might interfere with tokenization. Additionally,
Roberta, to evaluate its efficiency in handling large datasets. text is converted to lowercase uniformly to prevent
Harsha, T. M., Moukthika, G. S., Sai, D. S., Pravallika, M. N. fragmentation of words based on case variations.
R., Anamalamudi, S., & Enduri, M. [20] proposed an Tokenization, the segmentation of text into individual tokens
architecture of the project follows the MVC design pattern, or words, lays the groundwork for meaningful analysis by
with the code divided into different parts based on their structuring the text data. Stopwords removal comes next,
functionality. Lamba, D., Goyal, S., Chitresh, V., & Gupta, N. eliminating common and low-information words. Stemming
[21] introduced the Smart Resume Selector (SRS), an and lemmatization, other fundamental preprocessing steps,
integrated system designed for classifying resume categories standardize words by reducing them to their root form. These
based on resume-to-job matching. Srividya, K., Bommuluri, techniques enhance text consistency and ease further analysis.
S. K., Asapu, V. V. V. K., Illa, T. R., Basa, V. R., & Chatradi,
R. V. S. [22] introduced hybrid model merges extractive Feature engineering is paramount for advanced
summarization methods like Luhn and Textrank with the analysis, creating meaningful features or representations like
abstractive Pegasus model. Evaluation based on ROGUE TF-IDF vectors or word embeddings (e.g., Word2Vec,
scores compares its performance to that of BERT, XLNet, and GloVe). These capture word importance and semantic
GPT2, revealing enhanced outcomes compared to existing relationships between words. Addressing class imbalances is
models, thereby improving the summarization process's vital when they exist, as seen with job categories. Techniques
efficiency. like oversampling or under sampling balance the dataset to
mitigate class bias. Text length normalization ensures
uniformity, which some NLP models require. Data splitting
III. METHODOLOGY into training, validation, and test sets is standard practice for
model evaluation and preventing overfitting. Noise reduction,
The Fig. 3 shows the proposed method's framework. To including spell-checking and outlier removal, improves data
develop a comprehensive resume screening and job quality.
recommendation system that has advanced natural language
processing and graph neural network techniques to enhance In both datasets, text cleaning and normalization are
the efficiency and accuracy of the job matching process. This integral to creating a consistent and error-free dataset for
system aims to provide a tailored and data-driven approach to analysis. Moreover, in the context of resume classification,
assist both job seekers and recruiters by predicting job Part-of-Speech (POS) tagging plays a crucial role in the
categories, calculating matching scores, offering job and skill identification of words. POS tagging involves labelling words
recommendations, and continuously updating job listings. in a text corpus with their corresponding parts of speech, such
as nouns, verbs, adjectives, etc. In the context of the Resume
Dataset, POS tagging can assist in identifying and
A. Dataset Description categorizing words and phrases based on their roles within a
There are 2 different datasets used in this system and resume as shown in Fig. 1.
those datasets are collected from Kaggle. The first dataset,
referred to as the Resume Dataset, serves as the foundation
for a Graph Neural Network (GNN) model designed for
resume classification. This dataset contains 962 records with
two essential fields Category and Resumes. It is primarily
employed to predict job categories based on the provided

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 1041

Authorized licensed use limited to: Andhra University College of Engineering. Downloaded on February 18,2025 at 05:17:32 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

3. In GNN Model there are multiple layers, each responsible

for capturing and propagating information through a graph
structure. Starting from an initial node feature matrix, GNN
layers perform operations that combine node features, such as
weighted summation or attention mechanisms, to produce
updated node representations. This process is repeated over
multiple layers, enabling GNNs to capture complex
dependencies and patterns in graph-structured data effectively
Fig. 1. POS Tagging Example. and message passing between nodes is shown in Fig. 2.
When implementing a GNN model, these layers are
C. Model Implementation for predicting Job Role stacked sequentially to create the model architecture. The
After completing data preprocessing, the dataset is divided classification error, often measured using cross-entropy loss,
into training, testing, and validation subsets to facilitate the plays a pivotal role. During the training phase, the model's
resume screening procedure. Two models, namely K-Nearest parameters are fine-tuned to minimize this error. The process
Neighbours (KNN) and Support Vector Classification (SVC), involves iteratively adjusting the weights and biases of the
are trained on this meticulously prepared dataset. To optimize GNN through techniques like backpropagation and gradient
their performance, we conduct a thorough fine-tuning of the descent.
hyperparameters. However, the outcomes of these model
implementations do not yield substantial contextual insights
during the resume classification task. While these models do
exhibit proficiency in managing the pre-processed textual
data, they fall short when applied to unlabelled data, failing to
capture the nuanced contextual meaning present within the
text. This underscores the imperative need for more advanced
approaches in resume screening, ones that can accurately
decipher and categorize resumes based on their content.
The advanced deep learning model like GNN, has an
advantage of maintaining Contextual information in the data. Fig. 2. Message passing in GNN.
To implement GNN, for tokenizing and embedding the text
data, GloVe (Global Vectors for Word Representation) is
used. It loads pre-trained GloVe word vectors, which provide D. Score prediction between Job description and Resume
pre-trained word representations. An embedding matrix is
built using GloVe with capabilities for encoding, decoding, After data preprocessing and tokenization, the next
and embedding the text data into numerical formats, setting in the process is the calculation of the matching score between
the stage for the upcoming Graph Neural Network (GNN) a job description and a candidate's resume. The textual
methodology as shown in Fig. 3. This crucial component information needs to be transformed into a numerical format
operates on node embeddings and edge weights, facilitating to conduct this comparison. This conversion is accomplished
the flow of information and message-passing among nodes through the application of the Count Vectorizer from scikit-
within the graph. learn, transforming the text data into a document-term matrix.
This matrix serves as a representation of word frequency in
The TextLevelGNN Model serves as the core of the text
classification module within the GNN framework. The model each document, be it a resume or a job description.
receives inputs of node sets (text data), neighbour sets (graph Subsequently, the generated matrix becomes the foundation
connections), and public edge masks, establishing the for calculating the cosine similarity between these two
foundation for its message-passing operation is shown in Fig. documents.

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 1042

Authorized licensed use limited to: Andhra University College of Engineering. Downloaded on February 18,2025 at 05:17:32 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

Fig. 3. Resume classification and Job recommendation system using Graph Neural Network.

Cosine similarity is a powerful metric utilized in this In evaluating recommendation methods, the hybrid
context to gauge the angle between the term frequency vectors approach stands out as it combines collaborative filtering and
of the resume and job description in a multi-dimensional content-based filtering, offering a more comprehensive and
space. This percentage-based matching score reflects the personalized experience for users.
extent to which the candidate's qualifications match the job
requirements, offering valuable insights for recruiters and job
F. Web Scrapping Real Time Jobs
seekers is shown Fig. 3.
By talking job role as input latest jobs suitable for the
corresponding job role fetched from different websites using
E. Jobs and Skills Recommendation System web scraping. The data can be retrieved by initiating GET
The Job and Skill Recommendation System requests to the designated job URLs, granting access to the
represents a crucial component within the broader framework. most recent job postings. Following this, HTML parsing,
The process begins by taking the user's or job seeker's input facilitated by Beautiful Soup, structures, and navigates the
category, which signifies their specific domain or job type of web content, aiding in the precise identification of HTML
interest. It leverages a dataset of job listings augmented with elements housing the job listings. Once these listings are
key skills, encompassing essential details such as job titles, precisely located, the system extracts vital details,
associated key skills, and job categories as shown in Fig. 3. encompassing job titles, company names, locations, start
These are the foundational elements that feed the dates, and application links. This gathered data is then
recommendation engine. structured into dictionaries for each job listing and
amalgamated into a comprehensive list, thus delivering a real-
To assess the relevance of job listings and key skills time and organized compilation of job opportunities. This
to the user's input category, the system employs TF-IDF dynamic web scraping methodology empowers job seekers
vectorization. This technique converts textual data into with the most up-to-date job listings, furthering their pursuit
numerical vectors, effectively quantifying the importance of of meaningful employment prospects. Using Web
each term in relation to the entire dataset of job listings. scrapping we can recommend jobs using real time data that
Furthermore, the recommendation system extends its utility can be updated from different websites.
beyond job listings by identifying and suggesting relevant key
skills associated with the recommended job listings. To further
enhance the recommendation system's accuracy and IV. EXPERIMENTS AND RESULTS
personalization, a hybrid approach is introduced. This hybrid
recommendation system includes a variety of
The complete process, shown in Fig. 3, illustrates
recommendation methods, such as collaborative filtering and
the efficacy of the proposed method employing Graph Neural
content-based filtering. In this approach, user preferences,
Networks (GNN) in accurately predicting job roles. The
behaviors, and interactions within the platform can be factored
results of various models used in the system are comparison
into fine-tune recommendations. This not only empowers job
in Table 1. Performance metrics include precision, recall, f1-
seekers in discovering relevant job opportunities and
score, and accuracy are used for evaluating the system
understanding the key skills required in their chosen field but
performance. Accuracy serves as a pivotal performance metric
also assists employers in suggesting job titles and essential key
in evaluating the system's ability to precisely match job roles
skills.

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 1043

Authorized licensed use limited to: Andhra University College of Engineering. Downloaded on February 18,2025 at 05:17:32 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

to resumes, thereby affirming its effectiveness in enhancing time of the proposed algorithm is notably efficient, making it
the recruitment process. The results shown in Table 1 are well-suited for this application. Its swift processing ensures
proved the performance of GNN model. timely resume screening and job recommendations, a critical
factor in the dynamic job market, enhancing user experience
TABLE 1. COMPARISON OF PERFORMANCE METRICS FOR and job matching efficiency.
VARIOUS MODELS .
Performance metrics However, there is a scope for this work that lays the
Models foundation for a transformative path in the technology sector,
Precision Recall F1 – score Accuracy
providing students and job seekers with a novel approach to
KNN 0.9867 0.9745 0.9723 0.9745 streamline their job search. One crucial extension involves
SVM 0.9789 0.9612 0.9756 0.9712 integrating the system with direct outreach capabilities. After
accurately predicting job roles, users can select target
GNN 0.9886 0.9789 0.9799 0.9878
companies aligned with their career aspirations. The system
can then facilitate direct communication by generating
This high accuracy underscores the system's robustness in tailored email drafts addressed to the respective HR
categorizing resumes and job descriptions accurately for GNN representatives of these companies, enhancing both time
Model is shown in Fig. 4. Moreover, the system excels in efficiency and the effectiveness of job applications.
additional aspects as well. The scoring prediction module
ensures that matching scores between resumes and job REFERENCES
descriptions.
[1] Chung, Y. C., & Kuo, R. J. (2023). A domain adaptation approach for
resume classification using graph attention networks and natural
language processing. Knowledge-Based Systems, 266, 110364.
[2] Spoorthi, M., Kuppala, M., Karpe, V. S., & Dharavath, D. (2023,
March). Automated Resume Classification System Using Ensemble
Learning. In 2023 9th International Conference on Advanced
Computing and Communication Systems (ICACCS) (Vol. 1, pp. 1782-
1785). IEEE.
[3] Gu, Y., Wang, Y., Zhang, H. R., Wu, J., & Gu, X. (2023). Enhancing
Text Classification by Graph Neural Networks With Multi-Granular
Topic-Aware Graph. IEEE Access, 11, 20169-20183.
[4] Silva, G. L. L., Jayasinghe, T. L., Rangalla, R. H. M., Gunarathna, W.
K. L., & Tissera, W. (2022, December). An Automated System for
Employee Recruitment Management. In 2022 4th International
Conference on Advancements in Computing (ICAC) (pp. 346-351).
Fig. 4. Accuracy Graph for GNN Model. IEEE.
[5] Ransing, R., Mohan, A., Emberi, N. B., & Mahavarkar, K. (2021,
December). Screening and Ranking Resumes using Stacked Model. In
2021 5th International Conference on Electrical, Electronics,
Furthermore, the recommendation system leverages Communication, Computer Technologies and Optimization
resume content and predicted job categories to offer tailored Techniques (ICEECCOT) (pp. 643-648). IEEE.
suggestions for job roles and essential skills, enhancing the [6] Tripathi, R. C. (2022, December). An Approach of Intelligent
career development of job seekers. Additionally, the real-time Automated Resume Analysis & Recommendations. In 2022 Fourth
job web scraping component continuously updates the system International Conference on Emerging Research in Electronics,
with up-to-date job listings from various websites, Computer Science and Technology (ICERECT) (pp. 1-5). IEEE.
maintaining an enriched and current job database. This [7] Sridevi, G. M., & Suganthi, S. K. (2022). AI based suitability
measurement and prediction between job description and job seeker
guarantees that job seekers may access the most recent profiles. International Journal of Information Management Data
employment possibilities, supporting a successful and active Insights, 2(2), 100109.
job search process. [8] Wu, L., Qiu, Z., Zheng, Z., Zhu, H., & Chen, E. (2023). Exploring
Large Language Model for Graph Data Understanding in Online Job
Recommendations. arXiv preprint arXiv:2307.05722.
[9] Zheng, Z., Qiu, Z., Hu, X., Wu, L., Zhu, H., & Xiong, H. (2023).
V. CONCLUSION AND FUTURE SCOPE Generative Job Recommendations with Large Language Model. arXiv
preprint arXiv:2307.02157.
In summary, this approach introduces a robust [10] Li, S., Li, K., & Lu, H. (2023). National Origin Discrimination in Deep-
learning-powered Automated Resume Screening. arXiv preprint
resume screening and job recommendation system that arXiv:2307.08624.
effectively leverages advanced natural language processing [11] Zhu, Y. (2022, October). A Hybrid Job Recommendation Algorithm
and Graph Neural Networks (GNN). The system demonstrates for Intelligent Employment System Using User Profile-Based
exceptional prowess in predicting job categories, achieving a Filtering. In 2022 IEEE 2nd International Conference on Data Science
commendable 97% accuracy rate, and provides valuable and Computer Application (ICDSCA) (pp. 561-565). IEEE
insights for aligning resumes with job descriptions. The [12] Shaikym, A., Zhalgassova, Z., & Sadyk, U. (2023, June). Design and
Evaluation of a Personalized Job Recommendation System for
significance of this work lies in its potential to transform the Computer Science Students Using Hybrid Approach. In 2023 17th
job matching process, streamlining it and conserving time and International Conference on Electronics Computer and Computation
resources for both job seekers and employers. Nonetheless, it's (ICECCO) (pp. 1-7). IEEE.
important to acknowledge certain limitations in terms of [13] Sharma, M., Choudhary, G., & Susan, S. (2023, January). Resume
scalability and dealing with unstructured data. The execution Classification using Elite Bag-of-Words Approach. In 2023 5th

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 1044

Authorized licensed use limited to: Andhra University College of Engineering. Downloaded on February 18,2025 at 05:17:32 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

International Conference on Smart Systems and Inventive Technology Regex. In 2022 5th International Conference on Advanced Systems and
(ICSSIT) (pp. 1409-1413). IEEE. Emergent Technologies (IC_ASET) (pp. 44-48). IEEE.
[14] Petersheim, C., Lahey, J., Cherian, J., Pina, A., Alexander, G., & [19] Venkataramana, A., Srividya, K., & Cristin, R. (2022, October).
Hammond, T. (2022). Comparing Student and Recruiter Evaluations of Abstractive Text Summarization Using BART. In 2022 IEEE 2nd
Computer Science Resumes. IEEE Transactions on Education, 66(2), Mysore Sub Section International Conference (MysuruCon) (pp. 1-6).
130-138. IEEE.
[15] Du, Y., Luo, D., Yan, R., Liu, H., Song, Y., Zhu, H., & Zhang, J. [20] Harsha, T. M., Moukthika, G. S., Sai, D. S., Pravallika, M. N. R.,
(2023). Enhancing Job Recommendation through LLM-based Anamalamudi, S., & Enduri, M. (2022, April). Automated resume
Generative Adversarial Networks. arXiv preprint arXiv:2307.10747. screener using natural language processing (NLP). In 2022 6th
[16] Wosiak, A. (2021). Automated extraction of information from Polish international conference on trends in electronics and informatics
resume documents in the IT recruitment process. Procedia Computer (ICOEI) (pp. 1772-1777). IEEE.
Science, 192, 2432-2439. [21] Lamba, D., Goyal, S., Chitresh, V., & Gupta, N. (2020, May). An
[17] Pimpalkar, A., Lalwani, A., Chaudhari, R., Inshall, M., Dalwani, M., integrated system for occupational category classification based on
& Saluja, T. (2023, February). Job Applications Selection and resume and job matching. In Proceedings of the International
Identification: Study of Resumes with Natural Language Processing Conference on Innovative Computing & Communications (ICICC).
and Machine Learning. In 2023 IEEE International Students' [22] Srividya, K., Bommuluri, S. K., Asapu, V. V. V. K., Illa, T. R., Basa,
Conference on Electrical, Electronics and Computer Science V. R., & Chatradi, R. V. S. (2022, December). A Hybrid Approach for
(SCEECS) (pp. 1-5). IEEE. Automatic Text Summarization and Translation based On Luhn,
[18] Pant, D., Pokhrel, D., & Poudyal, P. (2022, March). Automatic Pegasus, and Textrank Algorithms. In 2022 International Conference
Software Engineering Position Resume Screening using Natural on Smart Generation Computing, Communication and Networking
Language Processing, Word Matching, Character Positioning, and (SMART GENCON) (pp. 1-8). IEEE.

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 1045

Authorized licensed use limited to: Andhra University College of Engineering. Downloaded on February 18,2025 at 05:17:32 UTC from IEEE Xplore. Restrictions apply.