Quasi-Metric Learning for Bilateral Person-Job Fit.

Area of Science:

• Computer Science
• Artificial Intelligence
• Information Retrieval

Background:

• Online recruitment relies on matching jobs with candidates, but existing methods struggle with asymmetric user preferences and feedback loops.
• Metric learning uses symmetric distances, failing to capture the two-way selection dynamics in recruitment.
• Current person-job fit models often use homogeneous graphs, overlooking the complex interactions between candidates and employers.

Purpose of the Study:

• To propose a quasi-metric learning framework for modeling asymmetric relationships in bilateral person-job fit.
• To develop a method that captures similarity propagation and users' two-way selection processes in online recruitment.
• To enhance the accuracy and effectiveness of job matching algorithms.

Main Methods:

• Developed a quasi-metric space satisfying triangle inequality for fine-grained similarity and incorporating asymmetric measures for user modeling.
• Constructed a heterogeneous relation graph of candidates, employers, and interactions.
• Proposed a relation-aware graph convolution network to capture mutual user effects through bilateral behaviors.

Main Results:

• The proposed quasi-metric learning framework effectively models asymmetric relations for bilateral person-job fit.
• The relation-aware graph convolution network successfully captures mutual effects in two-sided user interactions.
• Extensive experiments on real-world datasets validated the proposed methods' effectiveness.

Conclusions:

• The quasi-metric learning framework offers a theoretically sound approach to modeling recruitment dynamics from similarity and competitiveness perspectives.
• The integration of asymmetric measures and graph convolution networks significantly improves person-job fit in online recruitment.
• This work provides a novel and effective solution for complex bilateral matching problems in recruitment.