Prerna Job Connect: An AI-Driven Constraint-Aware Job Recommendation System for Women

Abstract

Existing works in the field of job recommendations mostly focus on either keyword matching or semantic similarity while failing to take account of any practical considerations including schedule, distance from the workplace, caregiver obligations, and mobility, among others. This often leads to recommendations of jobs that are compatible with user skills, but infeasible in practical use, especially for women. In order to fill this gap, the paper suggests a CARE-Net framework that integrates semantic matching of user skills and jobs along with evaluating practical constraints. The proposed recommendation system utilizes sentence embedding’s to represent skills of users and job demands, attention module to represent user constraints, dynamic weights learning technique to balance recommendation factors and SNAPY ranking function. This work was evaluated using the LinkedIn Job Postings Dataset and the performance was compared to BM25, SBERT, SBERT+Filter, and Attention-RS baselines. In particular, CARE-Net provided Precision of 0.847, Recall of 0.791, F1-score of 0.818 and NDCG of 0.863 while the constraint violation rate was reduced to 4.1% which is better than all other tested models.