You are on page 1of 6

JOB RECOMMENDATION SYSTEM USING

ENSEMBLE FILTERING METHOD


Preet Gada, Harsh Ranavat and Pranay Kathayat
Department of Computer Engineering
Shah And Anchor Kutchhi Engineering College
Mumbai, India
(preet.gada_2017, harsh.ranavat, pranay.kathayat) @sakec.ac.in

Abstract— - One of the most important aspects of the modern the time-consuming process of recruitment.
recruitment market is online job boards. With millions of job
seekers perusing job posts on a daily basis, the need for accurate, Candidate’s profiles are so diverse that it is laborious
effective, meaningful, and transparent employment suggestions is for recruiter to find the suitable competencies.
more important than ever. While recommendation systems have Consequently, it is important to identify the most
made significant progress in a variety of online areas by adding
social and economic value, the domain of job recommendation has important features of job candidature. . It also offers
received less attention. As a result, our goal is to create a software the capabilities of processing and analyzing large
application that can analyse a student's employability abilities. A quantities of unstructured job postings and job
job seeker must deal with a large number of data retrieval on the
internet and spend hours on the internet in order to discover the applications as to provide recruiters and job
proper match of a job. We established and created an online job applicants the ability to understand the preferences
seeking solution to help this challenging activity go faster and and requirements of each other. As a result, they can
smoother. Examining resumes and CVs is arduous work for
recruiters. To speed up the process of finding the suitable save time and efforts.
individual for the job.
II. LITERATURE REVIEW
I. INTRODUCTION In conducting our literature survey we were able
The expanding use of Internet has elevated the need to find significant research based on reading
for online employment chasing. As per Jobvite's different papers In our first paper “A Research
report 2014, 68% of on the web jobseekers are of Job Recommendation System Based on
school graduates or post graduates[4]. The key issue Collaborative Filtering”[1] Item based collaborative
is that a large portion of employment chasing sites filtering and log likelihood method was used. Item
simply show enrollment data to site watchers. based CF uses similarities between the rating
Understudies need to recover among all the data to patterns of items. We observed here finding similar
secure positions they need to apply. The entire items is easier. Attributes of items are more stable
method is dreary and wasteful. Job Recommendation than users’ preference, item-based methods are
work resides in the domain of online recommender suitable for off- line computing. Collaborative
systems and there are many algorithms to help a Filtering approaches often suffer from three
seeker find the right job, some are the traditional problems: cold start, scalability and sparsity In our
algorithms while some are founded in recent times second paper “Career Recommendation Systems
and there are a large number of hybrid algorithms using Content based Filtering”[2] Content based
which are a combination of many algorithms. The filtering method was used in this Users can give their
point of the current presentation is to give a feedback and criticism dependent on their
compelling technique for proposal for online experience on various parameters and The student’s
employment chasing. We trust to offering candidates profile can be handled in a more secure way by
a customized administration that can support them providing data encryption .We observed in this paper
secure ideal positions rapidly and helpfully In the that model can only make recommendations based
era of Big Data, both of employers and job seekers on existing interests of the user. In other words, the
are confronted with the increasing data overload and model has limited ability to expand on the users'
existing interests. In our third paper “A Graph-based Performance is decreased in user-item as matrix
Approach for Job Recommendation at Scale” Deep becomes bigger. It’s not scalable for huge data.
learning and graph based recommendation system Model-based methods are expensive models. Now
was used, Graph-based models methods addresses moving on to Content based filtering here the model
the shortcomings of CF-based approaches. Needs works well with implicit feedback while explicit
very thorough understanding of deep learning and rating is tough and its Quality improves over time as
Machine learning .This paper had the highest it is based on the likeness of the item by user. It has
accuracy but for this we need a high end GPU which the same issue as collaborative filtering of cold start.
isn’t available to me or my project mates. The graph- The performance here is limited to the traits applied
based approach addresses three main issues. Firstly in recommender system. Nowadays, automated
many of the parameters in the system were decided recommendation systems are already made which
analytically and with manual evaluation. For a are way better in performance wise than their
dynamic system, the researcher suggest that such counterparts. We found a lot of our paper and
parameters are better learned in a way to maximize important information in RecSys Challenge 2016,
user’s CTR (Click-Through-Rate). A Learning-To- which is an ensemble approach that combines both
Rank module could be utilized to learn to better sort collaborative filtering and content-based filtering.
recommendations based on user’s interactions with Ensemble methods mentioned above is basically a
previously served recommendations. Deep learning combination of multiple models usually produce
approaches pave the way towards language NLP more accurate results than a single model would. In
tools, so here in this paper the author is looking to recommender systems, all approaches have their
train more models to capture the similarity between certain strengths and weaknesses. So we make an
job postings for different skills and languages.[3] ensemble of CF and CBF methods finally. In order
to improve the application speed performance, The
In our fourth paper “, Job Recommendation based idea for our algorithm we got from RecSys
on Job Profile Clustering and Job Seeker Behavior” Challenge 2016, the approach of combination of
Cluster analysis approach was used this method collaborative filtering and content based filtering we
helps extract meaningful data from job posting propose for our application is not the best but rather
through and then job seekers are matched according a simple and easy solution. With in-depth
to their interactions. In this paper we observed that understanding of employment and requirement of
information is just available to the user has to find job hunters, the job recommender systems leave a lot
and match themselves. We read and observed many to be improved. To improve the quality of job
other papers but these Four were the ones that stood recommender systems, future work will concentrate
out and all 4 use different recommendation models on drawing a more detail users profile by history data.
and before stating our own application we wanted to
study all the different types of models out there.
III. OBJECTIVE
Another research paper we studied was “A survey of
job recommender systems” in this paper we got to Recommender systems aim to assist users in
know the main characteristics and challenges of finding their way through huge databases and
different recommendation systems For starters in catalogs, by filtering and suggesting relevant items
collaborative filtering, the characteristics are that it taking into account or inferring the users preferences
is Independent of any machine-readable (i.e., tastes, interests, or priorities). Study of
representation of the recommended material. It matching technique to match the best suitable
works smoothly with complex objects such as jobs candidate for the job position. Use of Collaborative
and movies. We don’t need to have a thourough Filtering to find the best suitable match. Use the
domain knowledge. The quality consistently content-based approach that takes into consideration
improves over time but we observed that there are an organization’s needs and the skills of the
various challenges faced as well like due to cold start candidate. Propose a test system that ranks students
it increases problems for new users and items.
in accordance with higher marks, therefore, to make
recruiters job easier..
IV. SCOPE
The main advantage of our app is the test-taking
feature we have included. When the candidate puts a
particular skill he/she has to give a test regarding that
skill if he/she passes only then it will be added so
because of this fake candidates gets eliminated this
feature increases our scope as in current scenario this
is not available in future we can also implement the
psychometric test which will further allow the
company to know the candidate better. So the scope
is very wide it depends on the development team
how far are they willing to go. On the other hand,
even if a person wishes to change his field we can 1(a): Collaborative filter flow
have jobs in the non-technical region as well.
V. ANALYSIS/ALGORITHM Linear regression-In statistics, linear regression
We are using both content based and collaborative is a linear approach to modelling the relationship
filtering for our recommender system between a scalar response and one or more
Collaborative filtering is a technique used by explanatory variables also known as dependent and
recommender systems. independent variables. The case of one explanatory
Collaborative filtering has two categories, a variable is called simple linear regression; for more
narrow one and a more general one. In the latest, than one, the process is called multiple linear
narrower sense, collaborative filtering is a method of regression. This term is different from multivariate
making automatic predictions (filtering) about the linear regression, where multiple correlated
interests of a user by collecting the weights of his dependent variables are predicted, instead of one
choices or get information from many users scalar variable.
(collaborating). The hidden assumption of the In linear regression, the relationships are modeled
collaborative filtering method is that if a person X using linear predictor functions whose unknown
has the same opinion as a person Y on an issue, X is model parameters are estimated from the
more likely to have Y's opinion on a different issue information. Such models are called linear models.
than that of a randomly chosen person. For example, Most commonly, the conditional mean of the
a shared television preferences program can predict response given the values of the explanatory
which television program a user should like given a variables or predictors is assumed to be an affine
list that is part of that user's likes (likes and dislikes). function of these values; less commonly, the
Note that these predictions are user-specific, but use conditional median or some another quantile is
information obtained from multiple users. This is utilized. Like every form of regression analysis,
different from the simple process of assigning an linear regression focuses on the conditional
(unspecified) rating to each item you like, for probability distribution of the answer given the
example based on the number of votes. values of the predictors, rather than on the joint
probability distribution of all of these cases and
variables, which is the domain of multivariate
analysis.
Content Based filtering-It is a machine learning
technology that uses the similarity of features to
make decisions. This technology is usually used in
recommendation systems, which are algorithms
designed to advertise or recommend things to users Model View Controller. Its main job is to have a
based on their accumulated knowledge.In this separate code base. Its purpose is to separate code
system, keywords are used to describe articles, and a and scope of responsibility in the software
user profile is created to indicate the types of articles development process. The main purpose of MVC is
the user likes. In other words, these algorithms try to to separate the project interface from the functions
recommend products that are similar to products that and data used in the application.
users liked in the past or are currently checking. It
does not rely on the user login mechanism to Dart is a programming language designed for
generate this usually temporary configuration file. In customer development, such as Web and mobile
particular, multiple candidate items are compared applications. It is developed by Google and can also
with items previously rated by the user, and the most be used to create desktop and server applications. It
matching item is recommended. This method is is an object-oriented, class-based garbage collection
rooted in information retrieval and information language with Cstyle grammar. Dart can compile
leakage investigation. native code and JavaScript. Support interfaces,
mixins, abstract classes, generic generics and type
inference.

Firebase is a platform developed by Google for the


creation of web and mobile applications. We used
firebase as our mobile application development
platform to create, improve, and develop our
application. We have used for its many features. As
firebase offers us a safe, secure and easy way to
manage our database this option for us was the best
available. The “build” group of products that we
used in development are:
[1] Authentication Firebase - Authentication takes
care of getting your users logged in and
identified. This product is essential to getting
some of the other products configured properly,
especially if you need to restrict access to per-
1(b): content based filter flow user data , which nearly every app will want to
do
Flutter is used for application development. Flutter [2] Cloud Firestore - Cloud Firestore Database
is an open source user interface software provide database services. I listed it as “realtime,
development kit where mobile app development is cloud hosted, NoSQL databases”. It likely
made more efficient it created by Google. We have addresses more of the needs and it’s also
used flutter to make an android application from a massively scalable.
single code base. Applications developed with [3] Cloud Storage - Cloud Storage provides
Flutter utilize the system features provided by the massively scalable file storage. It’s also
framework and are applicable to all libraries. This technically a Google Cloud product, not a
restrictive game plan can be easily manipulated to Firebase product. With Cloud Storage for
develop robust interface designs, and is a strong Firebase, you get client SDKs to use in your app
competitor to React Native in application that enable you to upload and download files
development services. Nevertheless, the flutter directly to and from your Cloud Storage
development service still has advantages in “bucket”.
application development, mainly due to the
appearance, execution and UI parts. MVC stands for
VI. IMPLEMENTATION
We have made a demo app to implement our idea.
It is just a prototype of our plan as many more
functionalities can be added to make it full-fledged
job recommendation portal. In our app first, you
enter your mobile number, you get the OTP. OTP
overcomes the disadvantages of the traditional
password-based authentication as here only the
valid person can log in as only he receives the one-
time password as a message on the number which
he has used to register. After you log in, you can
choose whether you are from the company side or a
job seeker and enter your basic details. From one 1.Company Portal
mobile number, you can register only once. If you If you are a job seeker then your dashboard looks
are a company, you will have to add your company like in the following images Depending upon your
details which will lead to your login and you will be skill sets you can apply to various jobs and in your
logged in the dashboard, later on even in the app edit profile you can update all your educational
you can easily navigate to company information details, you can also upload your resume by
and update it. There’s even the option to GST mandatorily has to be in pdf format and by clicking
number and industry type for the company as it will on badges you can give the quiz of any language
verify it is not a fake company. As you can see on such as C, Java, Python, etc and if you pass you get
the above snapshot the employer will be add his job a badge attached to your profile which will be
campaign where the employer can put in useful helpful in job selection.
details like job description, skills required,
experience in years, salary per annum, badges for
the skills and its level. After the job has been posted
the employer can view the campaign and see how
many users have applied for starters they will only
see the users name and its phone number but after
clicking on their name they will be able to see its
resume and important details of the user. This will
help the employer easily select the best candidates
and the candidates seen will already been filtered
according to their individual score making the job
of employer easier. On the profile side of the
company the employer will be able to access the 2.Student Portal
privacy policy, the terms and conditions and if they Moving on to our main feature of our application
have any queries the contact us option will lead the badge skill feature. As you can see you select
them the send query feature where the query will which badge the candidate wants to boast and for
saved in our firebase database and requested to badge to be shown in profile and to the employer
view upon the query. the candidate will have to go through a series of test
which determine the badge level.
[3] Walid Shalaby, BahaaEddin AlAila, “A
Graph-based Approach for Job Recommendation at
Scale” in IEEE, 2017
[4] Zheng Siting, Hong Wenxing,” Job
Recommender Systems: A Survey”in The 7th
International Conference on Computer Science &
Education, 2012.
[5] R. Moulouki, M. Y. El Ghoumari, ”Job
Recommendation based on Job Profile Clustering
and Job Seeker Behavior” in The International
Workshop on Artificial Intelligence, 2020.
[6] Nikolaos D. Almalis, Prof. George A.
Tsihrintzis” FoDRA - A New ContentBased Job
Recommendation Algorithm for Job Seeking and
Recruiting”
3. Test Feature [7] G. Shani and A. Gunawardana, "Evaluating
Recommendation Systems," Springer, pp. 257–298,
2011.
VII.CONCLUSION [8] D. H. Lee and P. Brusilovsky, “Fighting
So to conclude this project will help all the serious information overflow with personalized
candidates as all the fake and incapable people will comprehensive information access: A proactive job
not be present because of our test-taking feature. recommender,” in ICAS ’07: Proceedings of the
Also, we intend to implement the psychometric test Third International Conference on Autonomic and
which will be an additional feature to our app which Autonomous Systems, (Washington, DC,USA), p.
can be used by the company if they want. We intend 21, IEEE Computer Society, 2007.
to present a recommender system that aims to help [9] Y. Koren, R. Bell, and C. Volinsky, “Matrix
job seekers to find suitable jobs. First, job offers are factorization techniques for recommender systems,”
collected from job search websites then they are Computer, vol. 42, no. 8, 2009.
prepared to extract meaningful attributes such as job
titles and technical skills. Job offers with common
features are grouped into clusters. As a job seeker
like one job belonging to a cluster, he will probably
find other jobs in that cluster that he will like as well.
A list of top n recommendations is suggested after
matching data from job clusters and job seeker
behavior, which consists of user interactions such as
applications, likes, and ratings.

VIII. REFERENCES
[1] Yingya Zhang, “A Research of Job
Recommendation System Based on Collaborative
Filtering” in IEEE, 2014.
[2] Tanya Yadalam,Vaishnavi Gowda,” Career
Recommendation Systems using Content based
Filtering” in IEEE, 2020.

You might also like