Applying Knowledge Graphs to Recruitment: Construction, Tag Mining, and Recommendation at 58.com

The article introduces the business background of China’s recruitment market and 58.com’s role as a leading online hiring platform, highlighting the need for AI‑enabled solutions to improve bidirectional matching efficiency.

It then details the construction of a recruitment knowledge graph, which consists of three core tasks: building a multi‑dimensional, hierarchical tag system; mining tags from resumes and job postings using data‑augmentation techniques (Bootstrap rules, EDA, and DAGA); and extracting relationships between entities (candidates, positions, and companies) via pipeline and joint modeling approaches.

For tag mining, the authors describe challenges such as inconsistent labeling and domain‑specific terminology, and they present a three‑stage augmentation pipeline that triples the sample size and quintupled the tag count, leading to significant online performance gains.

The relation‑extraction section compares a traditional pipeline (entity extraction followed by relation extraction) with a joint SPO‑based method that uses a CNN+Attention backbone and pointer networks to handle overlapping relations, achieving higher F1 scores at the cost of increased inference time.

Application of the knowledge graph in recommendation is illustrated through several scenarios: reorganizing traffic with multi‑dimensional tags, balancing supply‑demand mismatches via graph‑based similarity search, and providing intelligent, step‑by‑step guidance to clarify user intent. These techniques have increased click‑through rates by ~4%, improved interview invitation rates by 2.5%, and lifted bidirectional connection rates by up to 5%.

Finally, the article outlines future directions, including further enrichment of the recruitment KG, extending its use to intelligent interview, QA, and credit‑risk assessment, and continuous optimization of personalized recommendation models.