Resume-Job_Description(CV-JD)-Matching

• Objective:

  • Build a PDF extractor to pull relevant details from CVs in PDF format, and match them against the job descriptions from the Hugging Face dataset.

• Dataset Used

  • Kaggle Resume Dataset
  • Job Descriptions from Hugging Face

• PDF Extractor(01_pdf-data-extraction.ipynb)

  • PDF Extraction was done using pdfplumber library

  • Python regex was used to extract Skills and Education Sections from the extracted text.

  • Finally the relevent information was stores in a csv file pdf_extracted_skills_education.csv

  • Challenges Faced: Using regex I could extract these Skills and Education part. But it was hard to generalise this over resumes of different format. So, more research will be needed to efficently extract these. Extracting Experience was really tough, I couldn't think of any regex that could extract Company_Name, Start_Date, End_Date within multiple headers(i.e., people with multiple experiences)

  • Propesed Solution: One thing that I know is training Custom Named Entity Recogniton(NER) but for this we need a custom tagged dataset for skills, Education, and Experience.

• CV-JD Matching(02_cv-jd-matching.ipynb)

  • Got the JD dataset from hugging face datasets library. 15 JDs from the dataset were selected for this project.
  • Basic text cleaning like lower_case, removing punctuations/emails/phone_numbers was done on the extracted resumes.
  • Tokenization and Embeddings for JDs & CVs were created using DistilBertTokenizer, DistilBertModel from transformers library.
  • For matching CV-JD, cosine_similarity was used from sklearn library.
  • Finally, Top-5 Candidates were extrated for the respective Job Descriptions, acoording to the respective similarity score.

• Overall Challenges Faced:

  • This was my first time working with PyTorch and transformers library
  • More than Modelling, Extracting the neccessary data is more tough as mentioned above in PDF Extractor.