nonameee12233/job-resume-matching

Resume–Job Matching Dataset

Dataset Summary

This dataset is designed for research and experimentation on resume-to-job matching, semantic retrieval, ranking, and recommendation systems. It contains candidate resume records, job posting records, and candidate-job compatibility records with multiple scoring features.

The dataset can be used for tasks such as:

• Resume-job matching
• Candidate ranking
• Job recommendation
• Semantic similarity modeling
• Skill-based matching
• Learning-to-rank experiments
• Graph-based recommendation models
• Embedding-based retrieval

Dataset Files

The dataset contains three CSV files:

File	Rows	Description
cv.csv	9,544	Candidate resume records with extracted profile information and text embeddings
job.csv	1,167	Job posting records with job descriptions, required skills, categories, and text embeddings
matches.csv	122,560	Candidate-job compatibility records with ranking labels and scoring components

Data Structure

cv.csv

Each row represents one candidate profile.

Column	Type	Description
candidate_id	string	Unique candidate identifier
source_row	integer	Source row index from the original resume data
target_position	string	Target or inferred job position for the candidate
resume_text	string	Full resume text
clean_skills	string	Cleaned skills represented as text
clean_skills_json	string	Cleaned skills in JSON-like format
education_text	string	Extracted education-related text
experience_text	string	Extracted experience-related text
years_experience	float	Estimated years of professional experience
inferred_category	string	Inferred job category for the candidate
embedding	string	Precomputed 768-dimensional text embedding

job.csv

Each row represents one job posting.

Column	Type	Description
job_id	integer	Unique job identifier
job_title	string	Job title
category	string	Job category
job_description	string	Original job description
job_text	string	Processed job text used for matching or modeling
clean_skills	string	Cleaned required skills represented as text
clean_skills_json	string	Cleaned required skills in JSON-like format
education_text	string	Education requirements or related text
years_required	float	Estimated years of experience required
embedding	string	Precomputed 768-dimensional text embedding

matches.csv

Each row represents a candidate-job compatibility record.

Column	Type	Description
candidate_id	string	Candidate identifier, linked to cv.csv
job_id	integer	Job identifier, linked to job.csv
final_score	float	Final normalized compatibility score
raw_score	float	Raw compatibility score before final normalization
label_strength	string	Match strength label: weak_positive, medium_positive, or strong_positive
semantic_score	float	Semantic similarity score between resume and job text
skill_score	float	Skill overlap or skill compatibility score
position_score	float	Position/title compatibility score
category_score	float	Category compatibility score
experience_score	float	Experience compatibility score
education_score	float	Education compatibility score
candidate_rank	integer	Rank of the job for the candidate
candidate_rank_pct	float	Percentile-style rank score for the candidate

Dataset Statistics

Overall Counts

Metric	Value
Candidate profiles	9,544
Job postings	1,167
Candidate-job compatibility records	122,560
Candidates appearing in compatibility records	9,532
Jobs appearing in compatibility records	1,164
Embedding dimension	768

Job Category Distribution

Category	Job Count
INFORMATION-TECHNOLOGY	240
BUSINESS-DEVELOPMENT	239
FINANCE	236
SALES	232
HR	220

Candidate Category Distribution

Category	Candidate Count
INFORMATION-TECHNOLOGY	8,140
FINANCE	1,010
SALES	215
BUSINESS-DEVELOPMENT	83
HR	80
UNKNOWN	16

Match Label Distribution

Label	Count
medium_positive	70,636
weak_positive	51,150
strong_positive	774

Score Ranges

Score Column	Min	Max	Mean
final_score	0.4300	0.7561	0.5857
raw_score	0.3060	0.6516	0.4359
semantic_score	0.5805	0.9177	0.7812
skill_score	0.0000	0.6170	0.0581
position_score	0.0229	1.0000	0.2656
category_score	0.0000	1.0000	0.8847
experience_score	0.0000	1.0000	0.6765
education_score	0.3000	1.0000	0.8974
candidate_rank_pct	0.3832	1.0000	0.9353

Intended Use

This dataset is suitable for building and evaluating matching and recommendation systems in the recruitment domain.

Example use cases include:

1. Semantic retrieval
   Use resume and job embeddings to retrieve relevant jobs for each candidate.

2. Learning-to-rank
   Train a ranking model using compatibility scores, ranking positions, and label strengths.

3. Classification
   Convert label_strength into classes for supervised classification.

4. Graph recommendation
   Build a heterogeneous or bipartite graph using candidates, jobs, and compatibility edges.

5. Hybrid matching models
   Combine semantic similarity, skill compatibility, position matching, category matching, experience matching, and education matching.

Example Usage

import pandas as pd

cv_df = pd.read_csv("cv.csv")
job_df = pd.read_csv("job.csv")
matches_df = pd.read_csv("matches.csv")

print(cv_df.shape)
print(job_df.shape)
print(matches_df.shape)

Load Embeddings

The embedding columns are stored as string representations of numeric lists. You can convert them back into arrays as follows:

import ast
import numpy as np

cv_df["embedding"] = cv_df["embedding"].apply(lambda x: np.array(ast.literal_eval(x), dtype=np.float32))
job_df["embedding"] = job_df["embedding"].apply(lambda x: np.array(ast.literal_eval(x), dtype=np.float32))

print(cv_df["embedding"].iloc[0].shape)
print(job_df["embedding"].iloc[0].shape)

Join Candidate, Job, and Match Data

merged_df = (
    matches_df
    .merge(cv_df, on="candidate_id", how="left")
    .merge(job_df, on="job_id", how="left", suffixes=("_candidate", "_job"))
)

merged_df.head()

Convert Match Labels to Numeric Classes

label_map = {
    "weak_positive": 0,
    "medium_positive": 1,
    "strong_positive": 2,
}

matches_df["label_id"] = matches_df["label_strength"].map(label_map)

Recommended Train/Validation/Test Splitting

For recommendation and ranking experiments, avoid random row-level splitting only. A row-level split may place records from the same candidate in multiple subsets and can overestimate model performance.

Recommended approaches:

• Split by candidate_id for candidate-level generalization.
• Split by job_id for job-level generalization.
• Use a time-based split if timestamp information is added in future versions.
• For graph-based models, ensure validation and test edges are not included in the training graph as target edges.

Data Quality Notes

• All columns in the provided CSV files are complete, with no missing values detected.
• Candidate and job embeddings are stored as text and should be parsed before numerical modeling.
• The dataset contains positive compatibility records. For binary classification or pairwise ranking, negative or unobserved candidate-job pairs may need to be sampled depending on the experiment design.
• Some candidates or jobs may not appear in matches.csv; this is expected when using a filtered compatibility table.

Ethical Considerations

Recruitment datasets should be handled carefully because automated matching systems can influence employment opportunities. Users of this dataset should:

• Evaluate models for bias across relevant groups where such metadata is available and appropriate.
• Avoid using protected attributes for ranking or filtering.
• Treat model scores as decision-support signals rather than final hiring decisions.
• Review recommendations with human oversight.
• Ensure compliance with applicable privacy, employment, and data protection regulations.

Limitations

• The dataset focuses on selected job categories and may not generalize to all occupations.
• Compatibility labels and scores should be interpreted as modeling signals, not definitive hiring decisions.
• The dataset does not guarantee complete coverage of every possible candidate-job pair.
• The embedding model used to create the stored vectors is not specified in the dataset files.
• Additional validation may be required before using models trained on this dataset in real recruitment workflows.

Citation

If you use this dataset in a project, report, or experiment, please cite the dataset repository.

@dataset{resume_job_matching_dataset,
  title  = {Resume-Job Matching Dataset},
  author = {Dataset Contributors},
  year   = {2026},
  url    = {https://huggingface.co/datasets/<your-username>/<your-dataset-name>}
}

License

Please update the dataset license before publishing. Use a license that matches your data source permissions and intended sharing policy.

Dataset Maintenance

For questions, updates, or corrections, please open an issue or discussion in the dataset repository.