Interpretable Job Recommendation with Reinforcement Learning and Talent Knowledge Graph Reasoning

Abstract: [Purpose] To address the limitations of existing job recommendation methods, including difficulty in large-scale application, cold start, lack of novelty, and interpretability, this paper proposes a reinforcement learning interpretable recommendation approach based on talent knowledge graph reasoning. [Method] A talent social experience knowledge graph is constructed based on a real resume dataset. Using reinforcement learning principles, a policy agent is trained on this knowledge graph, with the inference process divided into two sub-processes: selecting direction and selecting nodes. This enables the agent to identify potential high-quality recommendation targets within the knowledge graph. [Results] Compared to models such as LR, BPR, JRL-int, JRL-rep, and PGPR, the proposed reinforcement learning interpretable recommendation model demonstrates superior performance in terms of MRR@20 (81.7%), Hit@1 (74.8%), Hit@5 (92.2%), and Hit@10 (97.0%). [Limitations] The experimental dataset size and task type are relatively limited. [Conclusion] The model effectively integrates talent's historical work experience and similar talent's work experience for recommendations. By leveraging the attribute correlations of work positions within the knowledge graph, it provides reasoning paths along with recommendation results, effectively addressing cold start, lack of novelty, and interpretability issues.

Keywords: Job recommendation, knowledge graph reasoning, reinforcement learning interpretable recommendation.