BERTopic-LDA Model for a Bidirectional Recommendation System: Toward Matching Jobs and Job Seekers

The rapid development of online job portal system has promoted the use of recommendation systems to assist both job seekers and recruiters in the hiring potential candidate process. Current recommendation systems mainly deploy collaborative filtering or matching strategies. However, these models suffer from poor performance, data sparseness, scalability, cold start problems and lack the ability to offer explainable guidance for both employers and job seekers, Moreover to identify prospective candidate, then match it for opportunities, a diverse variety of knowledge and expertise is needed, although the conventional methods of recommendation based on matching are limited. The development of machine learning can made a significant impact overcome this problems and this limitations. natural language processing (NLP) applications frequently use topic modeling, a machine learning technique, to infer topics from unstructured textual data. Inspired by these recent progress and in order to assist both job seekers and recruiters, this article proposes a bidirectional job recommendation system to analyze job offers and profiles using the two topic modeling algorithms, BERTopic combined with Latent Dirichlet Allocation (LDA). We used job offer/resume dataset scraped from LinkedIn and job sites to evaluate the similarity of the bidirectional matching (employers and job seekers)