| dc.description.abstract | In today’s companies recruit a significant number of employees each year and face the
challenge of evaluating a vast quantity of resumes from candidates. Consequently, the HR
department finds it challenging to filter the CVs that best match the company's
requirements. This thesis addresses this challenge by rigorously comparing the accuracy
of an automated job-resume matching system against human-led decisions, specifically
focusing on filtering applicants based on their skill sets. The study employs preprocessing
techniques to standardize text and advanced Natural Language Processing (NLP)
techniques to extract skills from 122 potential resumes across 10 technical job positions.
Using Cosine similarity, Jaccard index, and Jaro-Winkler distance, the research analyzes
the similarity between job descriptions and resumes. The decisions made by the
automated system are then compared with the decision made by human assessors, and
their alignment is meticulously evaluated. Furthermore, leveraging K-means clustering,
the study categorizes resumes into "good," "average," and "poor" groups, providing
recruiters with efficient tools to identify the most qualified candidates. This research aims
to shed light on the strengths and limitations of machine-based decision-making in
recruitment, offering insights into enhancing efficiency, optimizing time usage, reducing
biases, and improving candidate evaluations. Through the evaluation of three similarity
measurement algorithms, the study identifies that Cosine similarity and Jaro-Winkler
distance provide high accuracy results, while the Jaccard index yields lower results. The
achieved research results contribute to a deeper understanding of automated systems'
efficacy in the recruitment landscape, leading to the development of more informed and
effective talent acquisition strategies, especially when compared to time-consuming and
biased manual recruiting processes. | en_US |
