ResuMatch – Resume Categorization

A Python-based application that automatically categorizes resumes into predefined job roles (e.g., Data Scientist, Java Developer, Business Analyst) using NLP techniques and machine learning models. Includes a Streamlit web interface for easy upload, bulk-processing, and CSV export of results.

---

🎬 Demo

---

Features

• Bulk Resume Upload: Upload multiple PDF resumes at once.
• Automated Text Extraction: Uses PyPDF2 to extract text from the first page of each PDF.
• Data Cleaning: Removes URLs, emails, special characters, and stop words via regular expressions and NLTK.
• Vectorization: Converts cleaned text into TF-IDF feature vectors.
• Multi-Class Classification: Trains and compares several classifiers (KNN, Logistic Regression, Random Forest, SVC, Multinomial NB, OneVsRest).
• Web Interface: Streamlit app for file upload, real-time categorization, and CSV download of results.
• Category-Based Storage: Organizes processed resumes into folders named after predicted roles.
• DOC-to-PDF Utility: Optional script to batch-convert .doc files to PDF.

---

Tech Stack

• Language: Python 3.x

• Libraries:

  • Data Processing & NLP: pandas, NumPy, re, nltk
  • Feature Extraction: scikit-learn (TF-IDF, LabelEncoder)
  • Machine Learning Models: scikit-learn (KNN, LogisticRegression, RandomForestClassifier, SVC, MultinomialNB, OneVsRestClassifier)
  • PDF Parsing: PyPDF2
  • Web App: streamlit
  • Model Serialization: pickle
  • DOC-to-PDF: docx2pdf (or equivalent)

• Environment Management: venv or conda

---

Dataset

• Source: Kaggle "Resume Dataset" ([link to dataset])
• Files:
  • updated_resume_dataset.csv (columns: category, resume)
  • resume_dataset.csv (alternative)
• Sample Size: ~960 resumes across ~12 job categories

---

Installation

1. Clone the repository:

   git clone https://github.com/namaniisc/ResuMatch.git
   cd resume-categorizer

2. Create and activate a virtual environment:

   python3 -m venv venv
   source venv/bin/activate      # On Windows: venv\\Scripts\\activate

3. Install dependencies:

   pip install -r requirements.txt

---

Usage

Jupyter Notebook

1. Launch Jupyter Lab/Notebook:

   jupyter notebook

2. Open resume_categorization.ipynb and run all cells:
   • Data loading & exploration
   • Visualization (bar plots & pie charts)
   • Text cleaning function (clean_text)
   • Encoding & TF-IDF vectorization
   • Model training & comparison
   • Save tfidf_vectorizer.pkl & model.pkl

Streamlit Web App

1. Ensure tfidf_vectorizer.pkl and model.pkl are in the root directory.
2. Run the Streamlit app:

   streamlit run app.py

3. In the browser:
   • Select one or more PDF resumes
   • Specify an output directory (default: categorized_resumes)
   • Click Categorize Resumes
   • Download the resulting CSV of filenames & predicted categories
   • Check categorized_resumes/<Category>/ folders for sorted PDFs

---

File Structure

resume-categorizer/
├── app.py                      # Streamlit web application
├── resume_categorization.ipynb # Jupyter notebook
├── model.pkl                   # Trained classification model (Logistic Regression)
├── tfidf_vectorizer.pkl        # Saved TF-IDF vectorizer
├── utils.py                    # Text cleaning & DOC-to-PDF functions
├── requirements.txt
└── README.md

---

Data Preprocessing

1. Cleaning: Removed URLs, emails, special characters.
2. Tokenization & Stop Word Removal: Using NLTK’s English stop word list.
3. Label Encoding: Converted job categories to numerical labels via LabelEncoder.
4. Vectorization: Applied TfidfVectorizer to convert text into feature vectors.

---

Model Training & Evaluation

• Split data: 80% train, 20% test (random_state=42).
• Baseline classifiers compared:
  • K-Nearest Neighbors
  • Logistic Regression (selected as best)
  • Random Forest
  • Support Vector Classifier
  • Multinomial Naïve Bayes
  • One-vs-Rest Logistic Regression
• Best Accuracy: ~99% on test set (Logistic Regression)

---

Saving & Loading Models

• Save:

  import pickle
  
  with open('tfidf_vectorizer.pkl', 'wb') as f:
      pickle.dump(tfidf_vectorizer, f)
  with open('model.pkl', 'wb') as f:
      pickle.dump(model, f)

• Load:

  with open('tfidf_vectorizer.pkl', 'rb') as f:
      tfidf_vectorizer = pickle.load(f)
  with open('model.pkl', 'rb') as f:
      model = pickle.load(f)

---

DOC-to-PDF Conversion

A utility function (convert_docs_to_pdf) uses docx2pdf to batch-convert .doc files in a directory to PDF:

from docx2pdf import convert
def convert_docs_to_pdf(input_dir: str):
    for filename in os.listdir(input_dir):
        if filename.endswith('.doc') or filename.endswith('.docx'):
            convert(os.path.join(input_dir, filename))

---

Future Improvements

• Integrate OCR for scanned PDF resumes.
• Add more advanced NLP (Named Entity Recognition).
• Deploy as a REST API (FastAPI / Flask).
• Add user authentication & dashboard.

---

Acknowledgements

• Kaggle Resume Dataset
• Streamlit Documentation
• Scikit-learn Documentation
• PyPDF2 & NLTK Libraries