Automated Resume Screening
COMP 4750: Natural Language Processing
Shawon Ibn Kamal
What is Resume Screening?
Resume screening is the process of determining whether a candidate is
qualified for a role based on their education, experience, and other
information captured on their resume.
The usual 3 steps to resume screening:
Minimum Preferred Shortlist for
Qualifications Qualifications Interview
Challenges to screen resume manually
The biggest challenge in screening resume is in handling the high volume of
applications received.
● On average, each corporate job offer attracts 250 resumes - upto 88% of
them tend to be unqualified.
● Of those candidates, 4 to 6 will get called for an interview, and only one
gets the job.
Statistics: Glassdoor
Challenges to screen resume manually
This high volume of resumes take up a lot of time for recruiters to make a
single hire.
● Average time-to-hire a new employee was 39 days in 2016, down from 43
days in 2015.
Source: Jobvite 2017 Recruiting Funnel Benchmark Report
More time spend on hiring implies higher cost.
● Average cost per hire statistics for companies is $4,129.
Source: SHRM Human Capital Benchmarking Report 2016
Automated Resume Screening
Goal: The goal of this project is to built an automated resume screening
application that tackles the challenges of screening manually
The application will consist of 2 main parts:
1. Parsing information from resume
2. Evaluation using ML and NLP
Parsing
● The first step to parsing is to extract text from PDF files. This step is quite
simple. We will do it using python pdfminer.
● The second step is to extract specific entities from resumes: names,
addresses, phone numbers, email addresses, education, experience, skills
etc.
○ This will be done using a NLP technique called Named Entity
Recognition (NER).
Parsing: Named Entity Recognition
● A NER model will identify noun phrase with the help of dependency
parsing and part of speech tagging (sometimes using CFG)
● Then, it will classify the extracted noun phrase into respective categories
using lookup tables and dictionaries
● To avoid misclassification, we can create a validation layer on top. Using
existing knowledge graphs can help get discrete result
Parsing: Difficulties
Evaluation
In this part of the application, we need to rank the resumes by its relevance to
the job description or requirements. It can be done using the ML process
called Information Retrieval.
There are many ways to rank documents, here we will use an unsupervised
method using vector space model.
The vectors representing the resume or query can be created using word2vec,
Bag of Words, Inverse Document Frequency etc. In this example, we will use
Inverse Document Frequency.
Evaluation
TF-IDF (term frequency-inverse document frequency) is a statistical measure
that evaluates how relevant a word is to a document in a collection of
documents. This is done by multiplying two metrics:
● how many times a word appears in a document, and
● the inverse document frequency of the word across a set of documents.
Evaluation
The vector space model will be based on similarity. Using cosine similarity,
we can calculate a similarity score for each resumes which is given below:
A -> vector of resume (document)
B -> vector of job description (query)
Evaluation
Evaluation: Other factors
Other factors to be considered while evaluating:
● Place of location
● Years of experience
● Relevance of previous experience
