Reproduce “ConFit: Improving Resume-Job Matching using Data Augmentation and Contrastive Learning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xiao Yu, Jinzhong Zhang, Zhou Yu

CONFIT combines paraphrase-based data augmentation with bidirectional contrastive learning to build independent dense embeddings for resumes and job postings. The resulting inner-product retrieval system improves ranking on two recruitment datasets, reaching up to 71.71 nDCG@10 for resume retrieval and 65.91 for job retrieval on AliYun, while supporting efficient FAISS-based search over unseen documents.

Key insights

  • Contrastive learning provides the main performance gain, while hard negatives increase the number and informativeness of training comparisons without requiring equivalent amounts of new labeling.
  • Combining EDA and ChatGPT paraphrases generally improves ranking, but augmenting every document is not consistently better than augmenting a selected subset.
  • Independent embeddings offer a practical trade-off: they enable millisecond-scale retrieval and generalization to unseen documents, though they cannot capture information absent from the resume or job posting.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.