Resume Generation System

Prem Borse1, Sahil Sinnarkar2, Ayush Sasane3, Suhasini Choudhari4, Amankumar Kumare5

Department of computer science engineering1, 2, 3, 4, 5, Savitribai Phule Pune university1, 2, 3, 4 ,5

Vishwakarma Institute of Technology Pune1, 2, 3, 4, 5
Maharashtra, India.

---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - We all know that job seeking is getting database. The resume sorting becomes more challenging
tougher day by day. As India annually produces 15 lakh when the job requirement demands a specialized skill set.
engineers graduates out of the which according to the In general, there is non-standard format to write the
survey of AICTE 60% engineering graduates passing out resume.
every year runs the risk of not getting a job at all. In such
scenarios, nobody can ever underestimate how important it 2. LITERATURE REVIEW
is to have a good screening process. A screening interview is
a type of interview that is conducted to determine if the The resume can be used for many reasons, but the
student has the qualifications needed to do the job for which main reason is used to secure employment. The
the company is recruiting. That’s why we developing a GUI resume is personal and academic information of
Based application for the resume generation and its employee, which a suitable employer sees related the
analysis. When a company is hiring an employee and job seeker and used to screen applicants often followed
requires the resume, the company receives thousands of by an interview. This project is deals with the parsing
resumes. The problem persisting here is, it’s very difficult for application developed for the resumes received
the Governing or the concerned authorities to go through through emails in multiple formats like Docx,
each and every parameter of the application and then select Document, text etc.
the candidates, for the interview. It’s a time-screening
process. The project will generate a resume for a candidate We are investigating the issues related to the
by concerning to the fix resume format and sort as per the development of approaches to improve the
requirement of recruiter or the concerned authority. performance of resume selection process. We have
extended the notion of special features and proposed
1. INTRODUCTION an approach to identify resumes with special skill
information. In the literature, the notions of special
Job market is getting tougher day by day. 20-33% out of features have been applied to improve the process of
the product selection in E- commerce environment. In this
1.5 million engineering graduates passing out every year, system, we have proposed an approach by considering
run the risk of not getting a job at all. In such scenarios, only skills related formation of the resumes. The
nobody can ever underestimate how important it is to have experimental results on the real-world data-set of
a good screening process. A screening interview is a type resumes show that the proposed approach has the
of job interview that is conducted to determine if the potential to improve the process of resume selection.
applicant has the qualifications needed to do the job for
which the company is hiring. A screening interview is We collected the resumes from the students of various
typically the first interview in the hiring process. Having a courses applying for a job. Students applying for the
successful screening process in place can make the job will send the resumes in different formats (doc.,
difference between finding the perfect fit for your team docx, pdf, text, etc.) and the information from all the
and making a bad hiring decision. resumes will be extracted in the database by using the
classification and regression techniques in data
In most cases, a screening interview includes a brief mining, and accordingly the students will be classified
review of your background, and a list of criteria designed according to their qualification. This project is a
to determine if you're a viable candidate for the position. A proposed model to find an appropriate evaluation
screening interview can be conducted over the phone or method for the classification of students and
in- person or simply going through resumes. A resume is a predicting the placement opportunity in an
summary of background, skills and qualifications, which is enterprise or a firm. This model will help the firm
sent to employers for review. The manager or concern to select the students in a convenient way to their
authority receives large numbers of Resumes from performance in the academics.
different sources which are difficult to process and store in
 have natural concurrency, as once a node is generated, all
3. PROBLEM STATEMENT of its children in the classification tree can be generated
concurrently. Furthermore, the computation for
When a company is giving a requirement and requires the
generating successors of a classification tree node can also
resumes, the company receives thousands of resumes. The
be decomposed by performing data decomposition on the
problem come up here is, it is very difficult for the
training data. Never- the less, parallelization of the
Manager or the concerned authorities to go through each
algorithms for construction the classification tree is
and every parameter of the resume and then select the
candidates, for the interview. It’s also a time-consuming challenging for the following reasons.
process. So, it is quite important that the applicant must have
a unique resume. 6. BACKGROUND AND INFORMATION
6.1 Support Vector Machines As one of the powerful
The proposed system is Time saving and very effective for techniques the SVM was first proposed by Vapnik [1] and
resume generation mechanism. It is easy for users as they recently has been extensively used in pattern classification.
just need to fill the information in given fix resume format on SVM maps the input dataset into a higher dimensional
portal. Resume can be converted to table format. Then feature space through several kernel functions. SVM then
resumes can be formatted as per the requirement of attempts to find the optimal separation hyperplane with the
particular authority. Also, personalization can be maximum margin in the feature space. Creation of a
performed if it is required. hyperplane wT x+w0=0 (w is the vector of coefficients and
w0 is a bias term) so that the optimal separating hyperplane
4.PROPOSED SYSTEM which has maximum margin linearly in the feature space is
find and can be posed as quadratic optimization problem.
Keeping all the above said problems in mind, we decided After finding the optimal hyperplane, the decision function is
to go one step further by developing the system which built as following: ( ) sgn[ ( , ) ] 1 f y y K x y b N i = ¦ i i i + = α
would assist extraction, storing and processing of resume θ Where N is the sample size, K (x , y) θ i represents defined
data by standardize the screening process helping in kernel function, θ is ҏa set of parameters and b is bias. α i are
increasing the accuracy and efficiency of whole creation non-negative Lagrange multipliers that satisfy ¦= = N i i i y 1
process. The systems allow user to go through several sets α 0 and are obtained by solving quadratic optimization
of questionaries for each and every parameter in it. The problem: i N i w α ¦ α = = 1 ( ) 2 1 ( , ) 1 1 ¦¦= = N i N j i j i j i y
system will allow user to filter through those parameters y K x y α α θ SVM for obtaining the optimal solution uses
according requirements and give the data extracted from different kernel function such as sigmoid, polynomial and
screening in standardize format which can be used in later radial basis function (RBF). Some kernel functions include
phases of recruitment process. polynomial, radial basis function (RBF) and sigmoid kernel,
The system contains three major modules: - in order to improve classification accuracy, these kernel
1. User portal for user to input their academic data parameters in the kernel functions should be properly set. In
2. System portal to extract data with help of various filters this study, RBF kernel function that is an effective option for
3. Output in standardize format for later recruitment classifying multi-dimensional data is applied. In the SVM
process. classifier with RBF kernel, the two parameters (C and Ȗ )
must be selected appropriately. 2009 Second International
5. METHODOLOGY Symposium on Knowledge Acquisition and Modeling 978-0-
7695-3888-4/09 $25.00 © 2009 IEEE DOI
In this module which use for taking the information that 10.1109/KAM.2009.86 17 2009 Second International
mean in this module student/user enter the all the Symposium on Knowledge Acquisition and Modeling 978-0-
information is based on two main algorithms: - 7695-3888-4/09 $26.00 © 2009 IEEE 6.2. Parameter
1. Decision tree Algorithm, Analysis in SVM The parameters in SVM such as C, Ȗ and the
2. Support Vector Machine Algorithm kernel function K(xi,y) exert a considerable influence on the
performance of SVM. Whether the value of C is too big or too
In the data mining domain, the data to be processed tends small can reduce the generalization of SVM. The value of C
to be very large. Hence, it is highly desirable to design indicates the error expectation in the classification process
Efficient as well as scalable algorithms. One way to of the sample data, and it affects the number of support
reduce the computational complexity of building a vectors generated by the classifier, thereby affects the
decision tree classifier using large training datasets is to generalization error of the classifier. If the value of x is too
use only a small sample of the training data. Such big, the separating error is high, the number of support
methods do not yield the same classification accuracy as vectors is small, and vice versa. The parameters in the kernel
a decision tree classifier that uses the entire data set [4, function reflect the characteristics of the training data, and
5]. In order to get reasonable accuracy in a reasonable they also affect the generalization of SVM. Therefore, only
amount of time, parallel algorithms may be required. after the choice of all these parameters are correctly made,
Classification decision tree construction algorithms can the SVM achieve its best performance. As to how to
select these parameters correctly, there is not a most
effective method up to now. Generally, the cross- by them.
verification trial or the gradient step-down operation is
adopted, but these methods involve too many human 8.REFERENCES
factors or the function is required continuously 1] N. Cristianini and J. Shawe-Taylor, “An Introduction to
differentiable, and the classifiers are susceptible to local SupportVector Machines and Other Kernel-based
minimum. 6.3 Artificial Immune System-Clonal Selection Learning Methods”, Cambridge University Press, 2000.
Theory In order to explain how an immune response is [2] L. N. De Castro and J. Timmis, “Artificial Immune
mounted when a non-self-antigenic pattern is recognized Systems: A New Computational Intelligence Approach”,
by a B cell, clonal selection theory is been developed. The Springer Verlag, New York, USA, 2002.
main goal of the immune system is to protect the human [3] De Castro, L. N. & Von Zuben, F. J. (2000), “The Clonal
body from the attack of foreign (harmful) organisms. The Selection Algorithm with Engineering Applications”,
immune system is capable of distinguishing between the GECCO’00 – Workshop Proceedings, pp. 36-37.
normal components of our organism and the foreign [4] Huang, C.-L., & Wang, C.-J. (2006). A GA-based feature
material that can cause us harm. These foreign organisms selection and parameters optimization for support vector
are called antigens. The molecules called antibodies play machines. Expert Systems with Applications, 31, pp. 231–
the main role on the immune system response. The immune 240.
response is specific to a certain foreign organism (antigen). [5] Hsu, C.-W., Chang, C.-C., & Lin, C.-J. (2003). A practical
When an antigen is detected, those antibodies that best guide to support vector classification. Technical report,
recognize an antigen will proliferate by cloning. This University of National Taiwan, Department of Computer
process is called clonal selection theory. The main features Science and Information Engineering, July, pp. 1–12
of the clonal selection theory are explored as follows [3]: [6] Hettich, S., Blake, C., & Merz, C. (1998). UCI repository
(1) generation of new random genetic changes of machine information and computer sciences, available
subsequently expressed as diverse antibody patterns by a at .
form of accelerated somatic mutation; (2) phenotypic
restriction and retention of one pattern to one
differentiated cell (clone); (3) proliferation and
differentiation on contact of cells with antigens. 2.4 Clonal
Selection Algorithm The clonal selection algorithm
developed on the basis of clonal selection theory of the
immune system. It was proved that can perform and adapt
to solve design of optimization tasks. In all runs of the
algorithm, the stopping criterion is a predefined maximum
number of generations. The clonal selection algorithm can
be described as follows (see Fig 1.) (1) Generate a set (P) of
candidate solutions, composed of the subset of memory
cells (M) added to the remaining (Pr) population (P = Pr +
M); (2) Determine (Select) the n best individuals of the
population (Pn), based on an affinity measure.

7.DISCUSSION AND CONCLUSION

In this paper, we present the results of using the clonal
selection algorithm to optimize the SVM parameters and
feature selection. By comparing the proposed clonal
selection algorithm with the standard genetic algorithm
and grid algorithm, we can notice that the clonal
selection algorithm can reach the optimal solution,
moreover, we proposed CSA strategy to select the
feature subset and set the parameters for SVM
classification simultaneous, and the experimental result
is better than GA approach. Results of this study are
obtained with an RBF kernel function. However, other
kernel parameters can also be optimized using the same
approach. Experimental results obtained from UCI
datasets, other public datasets and realworld problems
can be tested in the future to verify and extend this
approach. ACKNOWLEDGMENT The author would like
to thank Dr C.-J. Lin and C.-W. Hsu (National Taiwan
University, Taipei, ROC) that the support vector
machines were constructed with LIBSVM (Version 2.8)