Skip to content

the code of our paper "Beyond Matching: Modeling Two-Sided Multi-Behavioral Sequences For Dynamic Person-Job Fit" (实现十多个人岗匹配模型和动态人岗匹配模型的算法库,2021)

Notifications You must be signed in to change notification settings

BinFuPKU/DPJF-MBS

Repository files navigation

This work was done during my internship in Boss Zhipin, just be used for the purpose of academic exchange!

Beyond Matching: Modeling Two-Sided Multi-Behavioral Sequences For Dynamic Person-Job Fit

Paper: Bin Fu, Hongzhi Liu*, Yao Zhu, Yang Song, Tao Zhang, Zhonghai Wu*. Beyond Matching: Modeling Two-Sided Multi-Behavioral Sequences For Dynamic Person-Job Fit. The 26th International Conference on Database Systems for Advanced Applications (DASFAA 2021), April 11-14, 2021.

Abstract: Online recruitment aims to match right talents with right jobs (Person-Job Fit, PJF) online by satisfying the preferences of both persons (job seekers) and jobs (recruiters). Recently, some research tried to solve this problem by deep semantic matching of curriculum vitaes and job postings. But those static profiles don’t (fully) reflect users’ personalized preferences. In addition, most existing preference learning methods are based on users’ matching behaviors. However, matching behaviors are sparse due to the nature of PJF and not fine-grained enough to reflect users’ dynamic preferences. With going deep into the process of online PJF, we observed abundant auxiliary behaviors generated by both sides before achieving a matching, such as click, invite/apply and chat. To solve the above problems, we propose to collect and utilize these behaviors along the timeline to capture users’ dynamic preferences. We design a Dynamic Multi-Key Value Memory Network to capture users’ dynamic preferences from their multi-behavioral sequences. Furthermore, a Bilateral Cascade Multi-Task Learning framework is designed to transfer two-sided preferences learned from auxiliary behaviors to the matching task with consideration of their cascade relations. Offline experimental results on two real-world datasets show our method outperforms the state-of-the-art methods.

Running Requirements:

  • python 3.6+
  • pytorch
  • cuda10 + cudnn7 1.4.0

The source code is a demo used for academic exchange. Due to the company privacy policy, if you need this code for research (only for research), please contact with [email protected].

Bibtex:

@inproceedings{
  author    = {Bin Fu and
               Hongzhi Liu and
              Yao Zhu and
              Yang Song and
              Tao Zhang and
              Zhonghai Wu}
  title     = {Beyond Matching: Modeling Two-Sided Multi-Behavioral Sequences For Dynamic Person-Job Fit},
  booktitle = {The 26th International Conference on Database Systems for Advanced Applications (DASFAA 2021), April 11-14, 2021},
  year      = {2021}
}

About

the code of our paper "Beyond Matching: Modeling Two-Sided Multi-Behavioral Sequences For Dynamic Person-Job Fit" (实现十多个人岗匹配模型和动态人岗匹配模型的算法库,2021)

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages