Resume Scorer (Automated Resume Matching System)

This project implements an Automated Resume Matching System that matches resumes to job descriptions using machine learning models and various NLP techniques. It evaluates resumes based on multiple scoring methods such as TF-IDF, BERT embeddings, and machine learning classifiers like SVM and Naive Bayes, while also leveraging an Advanced Generative Model for content generation and evaluation.

Key Features

• Resume Parsing: Extracts text from PDF and DOCX files.
• Text Preprocessing: Cleans and transforms resume text into a usable format.
• Keyword Matching: Extracts and matches job description keywords with the resume using TF-IDF and BERT models.
• Multiple Scoring Models:
  • TF-IDF-based similarity score
  • BERT embeddings similarity score
  • SVM (Support Vector Machine) score
  • Naive Bayes score
  • Logistic Regression score
• Section Extraction: Identifies key sections in the resume like Skills, Education, and Experience.
• Link Validation: Checks if the URLs mentioned in the resume are valid and retrieves their content.
• Generative Model Integration: Uses an Advanced Generative Model to generate content and scores for advanced resume-job matching.
• Layout Analysis: Evaluates the resume's layout by checking for important sections using Optical Character Recognition (OCR).

Tools and Libraries Used

1. PDF and DOCX Parsing

• pdfplumber: Extracts text from PDF resumes.
• python-docx: Extracts text from DOCX resumes.

2. Natural Language Processing (NLP)

• nltk: Provides stopwords for text preprocessing.
• re (Regular Expressions): Cleans and preprocesses text data.
• TF-IDF: Implemented using TfidfVectorizer to extract key features from resumes.
• BERT (SentenceTransformer): Uses pre-trained BERT embeddings for advanced similarity matching between resumes and job descriptions.

3. Machine Learning Models

• SVM (Support Vector Classifier): Pre-trained SVM model (SVC_model.pkl) for evaluating resume match.
• Naive Bayes: Pre-trained Gaussian Naive Bayes model (GaussianNB_model.pkl).
• Logistic Regression: On-the-fly model trained using job description keywords.

4. Web Scraping and URL Validation

• requests: Retrieves content from URLs listed in resumes.
• BeautifulSoup (bs4): Extracts and cleans textual content from websites linked in resumes.

5. Optical Character Recognition (OCR)

• pdf2image: Converts PDF pages to images.
• pytesseract: Extracts text from images (PDF pages) for layout analysis.

6. Generative Model

• An Advanced Generative Model is used to generate content and evaluate the similarity between resumes and job descriptions, scoring each resume based on its content, layout, and structure.

7. Deployment and FastAPI

• FastAPI: Provides an API for uploading resumes, matching them with job descriptions, and returning a detailed evaluation.

Installation

1. Clone the repository:

   git clone https://github.com/DWE-CLOUD/Resume_Scorer
   cd Resume_Scorer

2. Install the required dependencies:

   pip install -r requirements.txt

3. Download the necessary NLTK data:

   python -m nltk.downloader stopwords

4. Ensure that Tesseract OCR is installed for layout analysis:

   • On Ubuntu:

     sudo apt install tesseract-ocr

   • On macOS (with Homebrew):

     brew install tesseract

Usage

API Endpoints

• Upload Resume and Match with Job Description: Upload a PDF or DOCX resume and provide a job description. The API will match the resume to the job description and return various scores:

  POST /upload/

  • Request Body:

    • file: The resume file (PDF or DOCX).
    • jd: The job description string.

  • Response: A JSON object containing:

    • ATS Score (TF-IDF): Score based on TF-IDF similarity.
    • BERT Score: Score using BERT embeddings.
    • SVM Score: Predicted score from the SVM model.
    • NB Score: Predicted score from the Naive Bayes model.
    • LR Score: Predicted score from the Logistic Regression model.
    • Layout Score: Score based on resume layout.
    • Links: Validated URLs mentioned in the resume.
    • Sections: Extracted sections like Skills, Experience, and Education.

• Root Endpoint:

  GET /

  Returns a welcome message.

Project Structure

• mdo/: Contains pre-trained machine learning models.
  • SVC_model.pkl: Pre-trained Support Vector Classifier model.
  • GaussianNB_model.pkl: Pre-trained Naive Bayes model.
• Notebooks/: Contains Notebooks for the training.
  • Start_main.ipnyb: Contains all the models and their first training and EDA.
  • Model_training.ipnyb: Contains the Models on 1000 vector features.
  • 5_feat_change.ipnyb: Contains the Models on 5 vector features.
  • 1000_to_5_change.ipnyb: Contains the Models on 1000 to 5 vector change features aka comparison.
• Dataset/: Contains Datasets.
  • URD.csv: Dataset of the Resume data.
• temp/: Contains a space to store resume
  • ----:
• requirements.txt: Dependencies required to run the project.
• main.py: Contains the FastAPI implementation.

Future Work

• Advanced Hyperparameter Tuning: Improve model performance by fine-tuning hyperparameters for SVM, Naive Bayes, and Logistic Regression.
• Deep Learning Models: Implement more advanced models like LSTM or Transformers for better resume-job matching.
• Extended Job Description Parsing: Parse complex job descriptions and categorize key requirements.