Design and Prototype Validation of a College Student Career Planning Agent Based on Large Language Models and Knowledge Graphs
DOI: https://doi.org/10.62517/jbdc.202601411
Author(s)
Mengdi Liu, Yubing Yang
Affiliation(s)
Ningbo University of Finance and Economics, Ningbo, Zhejiang, China
Abstract
In 2025, China produced 12.22 million university graduates. During the same year, intelligent manufacturing and artificial intelligence firms claimed that they were not able to fill technical positions. Both of these numbers are accurate and the distance between them is what this paper will be about. The career-planning products currently used do not serve students in any way that would be repeated in the market. A profile is created only once, when it is uploaded and never revisited. One suggested course comes without an explanation why, and thus neither student nor teacher can dispute the case. And the result is a document, which does not help the student on the next Monday. We constructed and experimented with an agent which separates the labour between a large language model (LLM), knowledge graph (KG) and state machine. The LLM reads unstructured text, the graph answers path questions which involve counting as opposed to guessing and the position of all learning tasks is held by the state machine in a loop we refer to as SkillTree so that progress is not lost due to page refresh. We experimented with three things and found one of them surprising. The most uninteresting is the parsing, which on 240 fields in 50 anonymised resumes has a full-field rate of accuracy at 87.5, with the dates and names of employers being the main source of error. On 30 annotated cases, matching was 95.0 percent of the expert ranking (which is not an interval), so it can be considered as a pilot and not a validation. It was the graph that came out to pay off: explainability before accuracy. There was a correlation between the scores of eight-dimension profile ratings by faculty members and the scores of the system r = 0.87. This test suite (219 cases) will cover about 90 per cent of core code and the concurrency of 500 threads will run without any request errors. Currently, there are only computer and information-technology jobs.
Keywords
New Quality Productive Forces; Career Planning; Large Language Model; Knowledge Graph; Person-Job Matching; Agent; Skill Tree
References
[1] Yue Changjun, Guan Yuting, Yan Siyu. An Analysis of the Employment Situation of College Graduates Nationwide in 2025. Journal of Higher Education, 2025, 46(11):9-24. (in Chinese)
[2] Chen Fangfang. Iterative Upgrading Path of College Students' Employment Guidance and Career Planning in the AI Era: A Cognitive Flexibility Perspective. Knowledge Economy, 2026(24):158-161. (in Chinese)
[3] Yang Miaomiao. Digital-Intelligence Empowerment of Career Planning and Employment Guidance Course Reform for Chemical Engineering Students. Chemical Fiber and Textile Technology, 2026, 55(7):201-203. (in Chinese)
[4] Xue Qian. Design of an AI-Based Career Planning Recommendation System for College Students. Heilongjiang Science, 2025, 15(1):1-5. (in Chinese)
[5] Zhang Guojun. Design of a Cloud Computing-Based Career Planning Recommendation System for College Students. Heilongjiang Science, 2025, 15(1):6-10. (in Chinese)
[6] Wu Yonghe, Jiang Yuanhao, Chen Yuanyuan, et al. Large Language Model-Supported Multi-Agent Systems: Technical Paths, Educational Applications and Future Prospects. Open Education Research, 2024, 30(5):63-75. (in Chinese)
[7] Z. Ji, N. Lee, R. Frieske, et al. Survey of Hallucination in Natural Language Generation. ACM Computing Surveys, 2023, 55(12):1-38.
[8] Zhang Lu, Peng Chengli, Du Yingte. Design and Evaluation of an Intelligent Question-Answering System for Intangible Cultural Heritage Communication in International Chinese Education: A Lightweight Path Coupling Knowledge Graphs with Large Language Models. Modern Educational Technology, 2026, 36(8):101-110. (in Chinese)
[9] Ding Ji, Cao Liping, Zhang Huan, et al. Optimization of a Knowledge Graph-Driven Collaborative Office System for Health E-Government: A Case Study of Cross-Department Official Document Intelligent Matching. Office Automation, 2026(19):11-13. (in Chinese)
[10] Sun Yusheng, Zeng Junhao, Chen Siyu, et al. Construction of a Knowledge Graph-Enhanced Policy LLM Question-Answering System. Research on Library Science, 2025(5):66-75. (in Chinese)