Information Retrieval and Extraction have wide ranging applications, from analyzing social media content and summarizing sports activities to other, more commercial applications like automation of customer service chats and search engines.We use our knowledge of the domain to solve this newly emerging problem of shortlisting candidates for a job. We provide the following functionalities: Extract important personal as well as professional details from the candidate resumes, presenting a more structured view of the provided information. We accept a ‘job description’, and then scan through the index of all the candidate resumes to dig out the 10 best candidates for the job. Besides, we also accept a resume, and present other resumes most similar to the given one. This helps if you know exactly what kind of person you need for the job. Moreover, we support ‘field queries’, whereby you may narrow down your search to incorporate specific details.

1. Job
Recommendation
Engine
Harshendra
Mahak Gambhir
Vishal Gupta

2. Introduction
Most job applications are through resumes
Short Listing resumes is a painful task
Especially when there are thousands of them
Manual checking is not feasible

3. Semantic Search
What is not a semantic search ?
Search based on keywords ?
Ex: grep “icpc”
Can we make it better ?
Weightage to sections
To be able to search in a particular section
Recommend resumes similar to a particular resume
NLP vs Natural Language Processing

4. Challenges
Structure of a resume
Content Segregation - How well is it written ?
Extraction of entities
Names
Marks/Grades
Previous Organizations
Years of experience
Skills

5. Problems addressed
Creation of info box for each resume.
Name: XYZ
Email : abc@gmail.com
Phone: xx-xxxxxxxxx
Skills
Projects
A search interface
Query> Machine Learning, ACM ICPC, Scikit Learn

6. The Process Flow
Resumes in
PDF format
PDF to text
Extract
features
Ranking
Model
Top K
Candidates
Resumes in
Text format
Infoboxes
JSON files
& Indexes
Used to display
a brief summary
of the candidate.

7. Approach
Each resume is a text document
Extract the sections in each resume
NER for names
Index the resumes according to sections
Extract features from each resume

8. Technical Details
Named Entity Recognition for extracting names
Used Stanford NER
Parse the sections using most common section names
Project/Projects/Academic Project/Major Projects/Minor Projects
Skills/Technical Skills/Technical Expertise and so on
Stanford temporal tagger for extracting the temporal
information
Jan 2012 - Jan 2014 (2 years)
Useful in extracting the years of experience

9. Technical Details
Bag of words approach
Vector space model
Extract feature for each resume
Boolean
A term is present or not d
Tf-Idf
Tf-idf weight of a term w.r.t the document d
Rank the resumes based on distance functions

10. Features
Let N be the |V|, d1 and d2 be two resumes
d1: v1 <v11 v12 . . . v1N>
d2: v2 <v21 v22 . . . v2N>
where,
vij: Tf-Idf(di,tj)
tj is the jth resume
Tf-Idf(d,t) = tft x log(N/dft)
tft - Term frequency of term t in document d
dft - document frequency of term t

11. Ranking
Convert the test query/document into tf-idf feature vector
t
Ex: test = “acm icpc, machine learning”
Calculate cosine similarity with each document and rank
them
cos-sim(d1,t), cos-sim(d2,t) and so on
Rank the resumes in non-decreasing order
Choose top K

12. Tools/Frameworks
Language: python 2.7
Frameworks:
Scikit-learn
Nltk
Stanford NER tagger
Stanford temporal tagger
Intuition!

13. Demo

14. Links
Code: https://github.com/gvishal/Sematic-Job-Recommendation-Engine
Deliverables: https://www.dropbox.com/home/IRE_Major_Project/deliverables
Presentation: http://www.slideshare.net/vishalg2/semantic-job-
recommendation-engine
Web Page: http://mahakgambhir.github.io/Sematic-Job-Recommendation-
Engine/
Video:
https://www.youtube.com/watch?v=EFiOd4-i2FM