Abstract:
"
The potential to use automation in the recruitment process is more appropriate for large companies,
which can revolutionize the recruitment process. Although the company has a high level of skilled
human resources team, it takes a lot of manpower, time and effort to select the best candidate for
an open position in any large company with the number of resumes received when advertised.
The recruitment process is now semi-automated. It's fair to believe that the first time you were
selected for a job vacancy in a company was a computer program, not people. The recruitment
process is a large process consisting of several stages. AI will be used to facilitate inefficient stages
of the recruitment process. It is simply the elimination of the human process of reading CVs.
Candidates are now tempted to include text in CVs using beautiful images, charts and fancy looks.
CVs can be sent in PDF type document to companies in various encodings such as UTF-8, ASCI,
Unicode. Existing solutions to automate the recruitment process are not a method of extracting text
from images and reading PDF documents with various encodings and extracting data from those
documents. These are the basic problems that this research hopes to solve. If semi-automated
software that can filter an applicant's CV cannot read CVs and extract critical information,
applicants will be excluded from the job competition. Every candidate has many factors to compare
among other candidates. Within them, Skills play an important role. So that Existing CV analysis
systems do not provide a CVs comparison process.
Combining OCR technology tools with natural language Processing presented a solution that could
extract text from images and read PDF documents with various encodings such as UTF-8, ASCI,
Unicode. The entire system was developed with a pre-evaluation by Domain experts. The system
stands out for its ability to read all encoded PDF formats and read text contained in images while
scaling with existing solutions. The system was evaluated and produced an accuracy of 92% and a
recall value of 82%, which was found to be very satisfactory in benchmarking similar solutions."
