A Text-Based Person-Job Matching Model Integrating Structured Features

In the recruitment industry, job recommendation algorithms have proven to be an effective solution to the problem of information overload. Specifically, person-job matching algorithms resolve job recommendation issues through bidirectional matching between job seekers and job descriptions. Most current person-job matching methods focus on non-structured text matching, which tends to neglect textual semantics and suffers from low efficiency. Additionally, when resumes are not well-structured or job seekers’ intentions are unclear, structured and semi-structured features become crucial in predicting matching accuracy. To address these challenges, this paper introduces a Multi-Feature Fusion Person-Job Matching Model (MFFPJM) that integrates both semi-structured and unstructured features to uncover latent correlations in resumes and job descriptions for a more effective representation of person-job match quality. The model utilizes BERT, BiLSTM, and attention mechanisms to learn the interactions across different representational subspaces of input features, establishing deeper connections between them. A multilayer perceptron is then used to predict the final match scores between resumes and job positions. The effectiveness of the MFFPJM model is demonstrated on a dataset released by Zhaopin.com, proving its capability in enhancing person-job match accuracy.