med2425/resume-job-fit-merged-v1

---
dataset_info:
features:
- name: resume
dtype: string
- name: jd
dtype: string
- name: label
dtype: string
- name: source
dtype: string
- name: resume_domain
dtype: string
- name: jd_domain
dtype: string
splits:
- name: train
num_bytes: 699473455
num_examples: 80017
- name: test
num_bytes: 125960894
num_examples: 13716
download_size: 300877211
dataset_size: 825434349
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
language:
- en
---

# 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

| Split | Examples |
|---------|----------|
| **train** | 80,017 |
| **test** | 13,716 |

**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

```python
from datasets import load_dataset

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

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

print(train[0])
```

## Citation
```bibtex
@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}}
}