Interpretable Job Recommendation Based on Talent Knowledge Graph Reasoning and Reinforcement Learning

Abstract: [Purpose] To address the limitations of existing job recommendation systems, such as difficulty in large-scale application, cold start, lack of novelty, and interpretability, we propose a reinforcement learning interpretable recommendation method based on talent knowledge graph reasoning. [Methods] We construct a talent social experience knowledge graph based on a real resume dataset and train a policy agent on the knowledge graph using reinforcement learning theory. The recommendation process is decomposed into two sub-processes: direction selection and node selection. This enables the agent to identify potential high-quality recommendation targets within the knowledge graph. [Results] Compared to LR, BPR.JRL-int, JRL-rep, and PGPR models, our reinforcement learning interpretable recommendation model based on talent knowledge graph reasoning achieves superior performance in MRR@20 (81.7%), Hit@1 (74.8%), Hit@5 (92.2%), and Hit@10 (97.0%). [Limitations] The experimental dataset size and task types are relatively limited. [Conclusion] Our model effectively integrates the historical work experience of talents with similar work experience and leverages attribute associations of job positions within the knowledge graph. It provides both recommendation results and reasoning paths, effectively addressing cold start, lack of novelty, and interpretability issues. [Keywords] job recommendation, knowledge graph reasoning, reinforcement learning interpretable recommendation.