Job Recommendation System based on Resume using Natural Language Processing and Distance-based Algorithm

This research aims to assist students struggling to determine which internship positions to apply for by providing recommendations based on their skills and abilities as stated in their CVs. Employing an experimental approach, key variables such as hard skills, soft skills, organizational experience, and job positions are extracted from student CVs, as well as job descriptions and requirements obtained from a corresponding job list. The extraction process is experimented with using OCR (Optical Character Recognition) and several PDF reader libraries. Through manual analysis, it is determined that the PDFPlumber library can handle layouts more effectively using character location data. After the extraction process, the variables obtained from the resumes are compared to those obtained from the job descriptions and requirements using distance-based algorithms. Three distance-based algorithms are utilized for this assessment. Results show that with Word2Vec vectorization, Euclidean distance yields the lowest average error value (0.051), followed by Jaccard distance (0.111) and Cosine similarity (0.26). Attempts to enhance Jaccard distance with synonims added did not yield significant improvements. Additionally, employing TFIDF for Cosine similarity improved performance, results in an error value of 0.117. It can be concluded, the Euclidean distance algorithm demonstrated the most effective performance.