Human-Post Intelligent Matching and Diagnosis System Based on Multi-Source Data Fusion and Knowledge Graph

DOI: https://doi.org/10.62517/jike.202604405

Author(s)

Wei Zhang1, Ge Gao1, Jinghan Yan1, Chenwei Zhu1, Zhen Hao1,2, Wanyue Liu1,2

Affiliation(s)

1School of Artificial Intelligence and Big Data, Henan University of Technology, Zhengzhou, Henan, China 2iFlytek Co., Ltd., Hefei, Anhui, China

Abstract

The iteration speed of job technical stacks in the new-generation information technology sector continues to accelerate, while recruitment job descriptions (JDs) commonly suffer from time-lag, plagiarism, noise, and skill inflation. Large language models tend to produce hallucinations when extracting job skill entities. Traditional static person-job matching methods fail to achieve dynamic evolution of job ability requirements and fine-grained competency gap diagnosis. To address these problems, this paper designs a multi-source heterogeneous data-driven dynamic post-ability knowledge graph and intelligent person-job diagnosis system. The system first integrates three types of heterogeneous data sources, namely crawled recruitment data, enterprise-published JDs, and parsed resumes, and designs a four-dimensional cross-validation mechanism for credible data cleaning and fusion. On this basis, a dual-agent pipeline consisting of an answer-generating agent and a hallucination-checking agent is constructed to suppress large-model hallucinations while supporting the dynamic identification of emerging skills and emerging posts. In addition, the Kahn topological sorting algorithm is introduced to generate personalized phased learning paths based on skill prerequisite dependencies. Experimental results on a self-built test set of 126 new-generation information technology JDs show that the JD parsing accuracy reaches 91.2%, resume skill extraction accuracy reaches 90.7%, and overall person-job matching accuracy reaches 90.3%. Ablation experiments further verify the individual effectiveness and synergistic gain of the four-dimensional cross-validation module and the dual-agent hallucination-checking module.

Keywords

Multi‑Source Heterogeneous Data; Post‑Ability Knowledge Graph; Large Language Model; Hallucination Suppression; Person‑Job Matching

References

[1]Kenthapadi K, Le B, Venkataraman G. Personalized job recommendation system at LinkedIn: Practical challenges and lessons learned//Proceedings of the 11th ACM Conference on Recommender Systems. 2017: 342-350. [2] Zhang Y S, Zheng J. Research on construction of post-skill knowledge graph for recruitment texts. Computer Engineering and Applications, 2023, 59(12): 231-240. [3]Lai R X, Zou S, Wu W H, Wu X Z. Review on application of knowledge graph in human-resource domain. Software Guide, 2021(6): 243-247. [4]Qin C, Zhu H, Xu T, et al. Enhancing person-job fit for talent recruitment: An ability-aware neural network approach//Proceedings of the 41st International ACM SIGIR Conference on Research and Development in Information Retrieval. 2018: 1205-1214. [5]Liu H, Wang S G. Review on hallucination-suppression techniques for large-language models. Computer Science, 2024, 51(4): 1-14. [6]Platanios E A, Sachan M, Neubig G, et al. Contextual parameter generation for universal neural machine translation//Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Brussels: Association for Computational Linguistics, 2018: 425-435. [7]Huang J C, Han T, Yi X Y. Socratic-debate multi-agent-based hallucination-suppression approach for large-language models. Expert Systems with Applications, 2026, 271: 132208. [8]Bourmpoulias S, Zeginis D, Tarabanis K. An entity-event knowledge graph for human resources management in public administration//2023 IEEE 25th Conference on Business Informatics. IEEE, 2023:1-8. [9]Zhang L, Liu Q. Research on adaptive-learning-path-generation method oriented toward smart education. Journal of Zhengzhou University (Natural Science Edition), 2022, 54(04): 102-108. [10]de Groot M, Schutte J, Graus D. Job posting‑enriched knowledge graph for skills‑based matching. arXiv preprint arXiv:2109.02554, 2021.