Job Recommendation Integrating Career Preference Diffusion and Person-Job Fit

Abstract: [Objective] To identify potential career preferences and reduce conflicts between recommended jobs and employment conditions, thereby improving job recommendation quality.
[Methods] Browsing, applications, and positive employer feedback are weighted to form direct preferences over occupational categories. A job-seeker association graph is constructed from job-seeker attributes and historical preferences, and finite-depth propagation supplements potential preference information. An aggregate constraint risk is calculated from location, education, work experience, and salary. This risk is combined with job-side statistical calibration to obtain a person-job fit score, which is integrated with the career preference score to generate recommendations.
[Results] Across five history-target holdout runs on the Zhaopin dataset, the proposed method improves NDCG@10 and MRR@10 by 13.6% and 21.1%, respectively, compared with LightGBM. Adding the person-job fit score produces small changes in relevance metrics and reduces the Top-10 strong-constraint violation rate by 29.5% compared with the model using career preference diffusion alone.
[Limitations] Interaction timestamps and their effects on changes in career preferences have not been considered.
[Conclusions] Career preference diffusion and person-job fit modeling provide complementary ranking signals, improving recommendation relevance while reducing conflicts with employment conditions.