Balancing Accuracy and Efficiency: A Comparative Analysis of Collaborative Filtering Algorithms for Job Recommendation Systems

Abstract

Recommender systems are commonly used to suggest relevant items to users, like movies or products. The digital transformation of the business sector has led to a surge in online job opportunities. This shift necessitates effective job recommendation systems to connect qualified candidates with relevant positions. This study evaluates the performance of four collaborative filtering algorithms for a job recommender system: Singular Value Decomposition (SVD),  SVD++ (SVDPP), co-clustering, and Non-Negative Matrix Factorization (NMF). We employ error rate, training time, and cross-validation performance as key evaluation metrics. Our findings reveal a trade-off between accuracy and efficiency. The co-clustering approach achieves the lowest error rates, indicating its effectiveness in recommending relevant jobs. However, this benefit potentially comes at the cost of increased training time compared to other methods. Conversely, the NMF-based model demonstrates significantly faster training times, making it computationally efficient.