med2425/resume-job-fit-merged-v1

Resume-Job Fit Dataset (Merged)

A high-quality dataset for training models to evaluate how well a resume fits a job description.

This dataset is designed for multi-class text classification (Good Fit / Potential Fit / No Fit).

Dataset Summary

Total: 93,733 examples

Features

• —`resume` (string): Complete resume text
• —`jd` (string): Complete job description text
• —`label` (string): Good Fit, Potential Fit, or No Fit
• —`source` (string): Origin of the example (ds1_original, generated_smart, synthetic_test, ...)
• —`resume_domain` (string): Detected domain of the resume
• —`jd_domain` (string): Detected domain of the job description

Data Generation Process

Training Set

• —Combined two public datasets: cnamuangtoun/resume-job-description-fit and kens1ang/resume-job-fit-augmented
• —Removed exact duplicates using MD5 hashing
• —Generated smart cross-domain pairs using domain classification and adjacency rules
• —Labeled using Qwen2.5-32B (via Ollama, temperature=0) as an expert recruiter

Test Set

• —Created from the original test split to prevent data leakage
• —Generated challenging synthetic pairs with the same domain logic
• —No text truncation applied
• —Independently labeled by Qwen2.5-32B

Label Definitions

• —Good Fit: Strong match in skills, experience, education, and role requirements.
• —Potential Fit: Partial match — candidate has potential but clear gaps exist.
• —No Fit: Significant mismatch in key requirements.

Domains Covered

software, data, ai, finance, marketing, healthcare, management, sales, design, hr, legal, engineering, other

Usage

from datasets import load_dataset

dataset = load_dataset("med2425/resume-job-fit-merged-v1")

train = dataset["train"]
test = dataset["test"]

print(train[0])

Citation

@misc{resume-job-fit-merged-v1,
  title = {Resume-Job Fit Dataset},
  author = {Mohamed Douali},                   
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/med2425/resume-job-fit-merged-v1}}
}