Dataset Card for Advanced Resume Parser & Job Matcher Resumes

This dataset contains a merged collection of real and synthetic resume data in JSON format. The resumes have been normalized to a common schema to facilitate the development of NLP models for candidate-job matching in the technical recruitment domain.

Dataset Details

Dataset Description

This dataset is a combined collection of real resumes and synthetically generated CVs.

• Curated by: datasetmaster
• Language(s): English
• License: MIT

Dataset Sources

• Real Resumes: Collected and normalized from various CV submissions.
• Synthetic Resumes: Generated using custom Python scripts leveraging Faker, with role-specific constraints.
• Repository: [Link to your GitHub repository or HuggingFace dataset page]
• Additional Resources: [Optional links to any papers or blogs describing the project]

Uses

Direct Use

• Training NLP Models: The dataset can be used to train and evaluate models for resume parsing and candidate-job matching.
• Data Augmentation: The synthetic resumes provide additional samples to augment limited real-world data.
• Exploratory Data Analysis (EDA): Researchers can analyze patterns in skills, education, and work experience among technical professionals.

Out-of-Scope Use

• Non-Technical Recruitment: This dataset is focused on technical roles and may not generalize well to non-technical industries.
• Personalized Hiring Decisions: The dataset is intended for model training and research purposes. It should not be used as the sole basis for hiring decisions without further validation.

Dataset Structure

Each resume in the dataset is stored in a nested JSON format with the following key sections:

• Personal Information: Contains name, contact details, location, summary, and social profiles.
• Experience: A list of work experiences, each with company information, job title, employment dates, responsibilities, and technical environment details.
• Education: A list of education entries including degree, institution, and relevant achievements.
• Skills: Detailed technical skills categorized by programming languages, frameworks, databases, cloud platforms, and additional languages.
• Projects: A list of projects showcasing practical applications of skills, including project description, technologies used, role, and impact.

Dataset Creation

Curation Rationale

The goal of creating this dataset was to build a comprehensive resource that combines both real and synthetic resume data. By merging these sources, the dataset provides a rich, diverse set of examples that are crucial for training robust NLP models for resume parsing and candidate-job matching.

Source Data

Data Collection and Processing

• Real Resumes:
  • Collected from CV submissions and pre-processed into a normalized JSON schema.
• Synthetic Resumes:
  • Generated using custom scripts with the Faker library, ensuring realistic and role-appropriate content.
  • Synthetic resumes were enriched with additional work experience, education, skills, and project details using predefined templates and random sampling.

Who are the Source Data Producers?

• Real Resumes:
  • Sourced from anonymized CV submissions provided during the project.
• Synthetic Resumes:
  • Generated programmatically to mimic real resume structures and content. They have been tailored to represent various technical roles based on industry-standard job keywords.

Annotations

This dataset does not include external annotations; all fields are either directly extracted from real resumes or generated synthetically.

Personal and Sensitive Information

The real resumes have been anonymized to remove personally identifiable information (PII). Synthetic resumes, generated by Faker, do not contain real personal data. However, some fields (e.g., name, email) are realistic placeholders.

Bias, Risks, and Limitations

• Bias:
  • The dataset focuses primarily on technical roles, which might not generalize to non-technical or creative industries.
  • Synthetic data may not capture all the nuances of real-world resumes.
• Risks:
  • Over-reliance on synthetic data might skew model training if not balanced with real data.
  • Incomplete fields in real resumes could introduce noise.
• Limitations:
  • Some fields (e.g., certifications, publications) are sparsely populated.
  • Variations in resume formats can result in inconsistent data quality.

Recommendations

• Balanced Training:
  • Combine real and synthetic data carefully to maintain diversity and realism.
• Data Cleaning:
  • Preprocess and impute missing values where necessary.
• Model Validation:
  • Validate models on an external dataset to mitigate overfitting to synthetic patterns.

Citation [optional]

If you reference this dataset in academic work, please cite it as follows:

BibTeX:

@misc{your_dataset2023,
  author={datasetmaster},
  year={2023},
  howpublished={\url{https://huggingface.co/datasets/datasetmaster/resumes}},
}

APA:

[Dataset]. HuggingFace. https://huggingface.co/datasets/datasetmaster/resumes

Glossary [optional]

• Resume Parser: A system that extracts structured information from unstructured resume documents.
• Synthetic Data: Artificially generated data that simulates real-world data.
• NLP: Natural Language Processing, a field of AI focused on the interaction between computers and human language.

More Information

For further details on the dataset creation process and intended uses, please refer to our project documentation