Hybrid Job Recommendation Model Integrating TF-IDF, SBERT, and Weighted Scoring System

Resumo

The recent breakthroughs, particularly transformer-based models brought a significant improvement in job recommendation systems. These models are semantic-based. However, this type of approach can still be improved. The proposed hybrid job recommendation model integrates Term Frequency-Inverse Document Frequency (TF–IDF), Sentence Bidirectional Encoder Representations Transformers (SBERT) model, and a weighted scoring system, which comprises a semantic boost and skill gap penalty to achieve a more comprehensive and contextual recommendation process. In contrast to models that treat all candidate features uniformly, this system assigns distinct weights to attributes like skills, experience, and education through a weighted scoring system. It also integrates a semantic boost, which considers the clustering of terms and recency of expertise. The hybrid job recommendation system also includes the skill gap penalty, which is used to reduce scores when essential skills listed in a job description are missing from the candidate’s profile. The 1.0 score of both Normalized Discounted Cumulative Gain (NDCG) and Mean Average Precision (MAP) signifies that the scoring underlying features and the logic used are highly effective in identifying and prioritizing relevant job opportunities for your profile within this dataset.