Multilingual Skill Matching System for Freelancers

This document is a research article detailing a novel neural retriever architecture designed for efficient skill matching between freelancers and job proposals in a multilingual context. It addresses the challenges of aligning diverse freelancer profiles with project descriptions, focusing on scalability and the effective utilization of comprehensive profile information. The proposed method employs pre-trained multilingual language models as the backbone of a custom transformer architecture, allowing for the encoding of both project descriptions and freelancer profiles. The architecture is specifically trained using a contrastive loss on historical data to optimize the similarity in skill matching. Various sections of the article present the architecture and retrieval model, outline prior research limitations, detail the experimental protocol, and discuss the deployment results. Additionally, the document elaborates on prior approaches in the HR domain, elucidating the advancements in person-job fit algorithms through enhanced matching processes.