# Generative Job Recommendations with Large Language Model

**Authors:** Zhi Zheng, Zhaopeng Qiu, Xiao Hu, Likang Wu, Hengshu Zhu, Hui Xiong  
**arXiv:** [2307.02157](https://arxiv.org/abs/2307.02157)  
**Date:** 2023-07-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2307.02157v1/

## Key points

- Rapid development of online recruitment services
- Use of recommender systems to simplify job seeking process
- Current job recommendation models rely on collaborative filtering or person-job matching strategies
- Lack ability to provide explainable guidance to job seekers
- Limited potential as comprehensive career AI advisors
- Proposal of GIRL (GeneratIve job Recommendation based on Large language models)
- Supervised Fine-Tuning (SFT) strategy used to train LLM-based generator for crafting suitable Job Descriptions (JDs) based on CVs
- Training a model to evaluate degree of matching between CVs and JDs as reward model
- Proximal Policy Optimization (PPO)-based Reinforcement Learning (RL) used for further fine-tuning the generator based on recruiter feedback
- GIRL serves as a job seeker-centric generative model that provides personalized suggestions without relying on a candidate set
- Enhances performance of existing job recommendation models by supplementing job-seeking features with generated content
- Extensive experiments conducted on large-scale real-world dataset demonstrate effectiveness of GIRL
- Paradigm-shifting approach leveraging large language models in job recommendation systems
- Aims to foster improved and more tailored job-seeking experience for individuals.

## Summary

The rapid development of online recruitment services has led to the use of recommender systems to simplify the job seeking process. However, current job recommendation models primarily rely on collaborative filtering or person-job matching strategies and often operate as "black-box" systems. They lack the ability to provide explainable guidance to job seekers and are limited in their potential as comprehensive career AI advisors. To address these limitations, the authors propose GIRL (GeneratIve job Recommendation based on Large language models), a novel approach inspired by recent advancements in Large Language Models (LLMs). The authors initially employ a Supervised Fine-Tuning (SFT) strategy to train an LLM-based generator for crafting suitable Job Descriptions (JDs) based on a job seeker's Curriculum Vitae (CV). They also propose training a model that can evaluate the degree of matching between CVs and JDs as a reward model. Proximal Policy Optimization (PPO)-based Reinforcement Learning (RL) is then used to further fine-tune the generator based on recruiter feedback, tailoring the output to better meet employer preferences. Unlike traditional job recommendation models, GIRL serves as a job seeker-centric generative model that provides personalized suggestions without relying on a candidate set. This capability enhances the performance of existing job recommendation models by supplementing job-seeking features with generated content. Extensive experiments conducted on a large-scale real-world dataset demonstrate the substantial effectiveness of GIRL. Overall, this research introduces a paradigm-shifting approach to job recommendation systems by leveraging large language models. By offering personalized and comprehensive recommendations, GIRL aims to foster an improved and more tailored job-seeking experience for individuals.