Generative Job Recommendations with Large Language Model (2307.02157v1)

Abstract: The rapid development of online recruitment services has encouraged the utilization of recommender systems to streamline the job seeking process. Predominantly, current job recommendations deploy either collaborative filtering or person-job matching strategies. However, these models tend to operate as "black-box" systems and lack the capacity to offer explainable guidance to job seekers. Moreover, conventional matching-based recommendation methods are limited to retrieving and ranking existing jobs in the database, restricting their potential as comprehensive career AI advisors. To this end, here we present GIRL (GeneratIve job Recommendation based on LLMs), a novel approach inspired by recent advancements in the field of LLMs. We initially employ a Supervised Fine-Tuning (SFT) strategy to instruct the LLM-based generator in crafting suitable Job Descriptions (JDs) based on the Curriculum Vitae (CV) of a job seeker. Moreover, we propose to train a model which can evaluate the matching degree between CVs and JDs as a reward model, and we use Proximal Policy Optimization (PPO)-based Reinforcement Learning (RL) method to further fine-tine the generator. This aligns the generator with recruiter feedback, tailoring the output to better meet employer preferences. In particular, GIRL serves as a job seeker-centric generative model, providing job suggestions without the need of a candidate set. This capability also enhances the performance of existing job recommendation models by supplementing job seeking features with generated content. With extensive experiments on a large-scale real-world dataset, we demonstrate the substantial effectiveness of our approach. We believe that GIRL introduces a paradigm-shifting approach to job recommendation systems, fostering a more personalized and comprehensive job-seeking experience.

• The paper introduces GIRL, a framework that generates custom job descriptions from CVs using LLMs.
• It employs supervised fine-tuning, reward model training, and reinforcement learning to align job recommendations with recruiter feedback and market needs.
• GIRL improves recommendation performance in cold-start scenarios while enhancing system explainability and user trust.

Generative Job Recommendations with LLMs

The paper "Generative Job Recommendations with LLM" investigates a novel paradigm for job recommendation systems leveraging the capabilities of LLMs. It introduces GIRL (GeneratIve job Recommendation based on LLMs), which generates personalized job descriptions (JDs) for job seekers by analyzing their curriculum vitae (CVs). GIRL departs from traditional recommender systems that primarily focus on scoring and ranking existing jobs. Instead, GIRL explicitly creates new job descriptions, providing more comprehensive career guidance and enhancing system explainability.

Existing Challenges in Job Recommendation Systems

Existing job recommendation systems employ methods such as collaborative filtering and person-job matching. These methods rely heavily on neural networks to produce matching scores based on specific job seeker and job information. Their primary challenge lies in the black-box nature of these models, which hampers explainability and user trust. Furthermore, traditional models are limited to retrieving and ranking pre-existing jobs, thus failing to provide original, tailored job suggestions for individual job seekers.

Traditionally, the significant semantic gap between CVs and JDs further impacts the performance of these systems, leaving a substantial room for improvement in aligning job recommendations with both job seekers' profiles and market demands.

Figure 1: Schematic diagram of three distinct job recommendation paradigms.

GIRL: A Generative Approach Leveraging LLMs

The primary contribution of this work is the introduction of a paradigm-shifting generative approach for job recommendations using LLMs. The GIRL framework uses a three-step training methodology that harnesses the sophisticated generation capabilities of LLMs to generate personalized job descriptions:

1. Supervised Fine-Tuning (SFT): The initial stage involves training the LLM-based generator on matched CV-JD pairs to learn how to construct appropriate job descriptions from CV data using a tailored prompt template (Figure 2).
2. Reward Model Training (RMT): A reward model is trained to capture recruiter feedback, distinguishing between well-matched and mismatched CV-JD pairs. It mimics real recruiter decisions and uses these insights to adjust the generative process.
3. Reinforcement Learning from Recruiter Feedback (RLRF): The fine-tunning of the generator via Proximal Policy Optimization (PPO) enables it to better align with recruiter preferences, effectively incorporating market demands into the entries generated by the LLMs.

   Figure 3: The training framework of the generative recommendation model.

Evaluation and Performance

The paper evaluates GIRL's performance across two main axes: the quality of generated job descriptions and the impact of these descriptions on enhancing traditional recommendation tasks.

1. Generation Quality: The GIRL model, with its combination of SFT and RLRF, consistently produces more detailed and relevant job descriptions than competitor LLMs, as validated by ChatGPT-based evaluations (Figure 4). Specifically, GIRL demonstrates superior capability in aligning job seeker traits with JD content.
2. Recommendation System Enhancement: Leveraging the generated JDs as an augmentation to job seeker profiles, GIRL significantly improves the performance metrics of existing discriminative recommender models. This enhancement is particularly evident under cold-start conditions where traditional models typically struggle.

   Figure 5: Performance of different models with different generation number.

Implications and Future Directions

The successful integration of generative approaches in job recommendation paradigms marks a critical shift away from merely reactive systems toward proactive career guidance solutions. These innovations not only extend the functional capabilities of recommender systems but also enhance the overall user experience by ensuring greater alignment with individual needs and evolving market requirements.

Looking forward, further advancements may focus on expanding the applicability of such systems to different domains and employing multi-modal data to fine-tune the recommendations. Additionally, the integration of more complex socio-economic indicators into the reward model could further contextualize and personalize job recommendations.

Conclusion

The GIRL framework represents a significant step forward in the evolution of job recommendation systems. By leveraging the generative capabilities of LLMs, the model addresses several long-standing challenges in this domain, including user trust and system explainability, while concurrently presenting significant opportunities for enhancing the accuracy and relevance of job recommendations in digital hiring platforms. As such, GIRL serves as a foundational approach for future exploration and development within AI-driven recommendation systems.