- Extract the important information from the resume
- Match the resume to the provided expertise list (This list was provided by Vinod to Harsha)
- Match the resume to the Job description (For Concept purpose, I have downloaded the data engineer resume from the job board of travelers insurance)
- Execute the below command from your terminal for installing all the required libraries
pip install requirement.txt
- Download the spacy word embedding model using the below command
python -m spacy download en_core_web_md
python resumeparser [-resume resumeFile]
-resume; input resume file
Example:
python resumeparser -resume "data/resumes/resume_trail4.pdf"
when you execute the code, the code pushes the output to the respective columns in the database and prints the contents in the json format
- Download the DB Browser from here for viewing the sqlite database
-
Mobile, Email Address, Github, Linkedin
- Used Regex
-
Address
- Used the combination of Regex and pyad library (for standardization)
-
Name
- Used the chunking and chinking concepts in combination with a lookup table. Unlike any NER model, this process can prescribe the name rather than predict
This process is proven to give better results rather than using any pretrained NER models as its hard for any NER model to understand the variations and the variety of resume.
- Used the chunking and chinking concepts in combination with a lookup table. Unlike any NER model, this process can prescribe the name rather than predict
- For every word in the expertise list, get the glove word embedding provided by spacy library
- Then get the similarity score of the each word embedding to all the
nouns(do the parts of speech tagging and eliminate all other part of speech words other than nouns) of the resume; - The top 5 highest similarity score words are the top 5 as the expertise of the resume
We have used the pretrained glove embedded model, in an ideal scenario we need to train our
own word embedded model on a specfic domain corpus. But the results from the above approach
seem compelling too.
Examples:
--------------------------------------------------------------------------------------------
-'resume_trail4'(data/resumes) is the resume of an Orthopaedic Spine Surgery, below are the top5 expertises
the code matched from the expertise list
'radiology, internal medicine, general dentistry, general surgeon, physical therapy'
--------------------------------------------------------------------------------------------
- 'resume_trail2' is the resume of a Building inspector (related to construction), below are
the top5 expertises the code matched from the expertise list
'highway design & construction, public health, life care planner, fire cause & analysis,
hospital administration'
--------------------------------------------------------------------------------------------
- Get the Glove word embedding to every
noun and verbof the resume and the job description. - The calcuate the similarity score between those documents with only verbs and nouns
All the traditional ATS systems uses the keywords to filter out the resumes, however there are
many scenarios where keyword approach is proven wrong. This approach is presented as a POC how can we
better construct the ATS system with word embedding as they care both for keywords and the dependency
between words
Examples:
Currently the job description to be matched is a data engineer position from Travelers Insurance
Similarity score between Resume_Harsha_AIScientist (my resume) and job description is 0.90
Similarity score between resume_trail2 (related to construction) and job description is 0.71
We can clearly see how this process can be efficient in picking the best resumes from a pool of
resumes.