Resume parser Solution for Code4Goal – Coding Contest Authored and maintained by Lizurchik Alexey, 2015 The Problem Often Companies have problems with sorting out large volumes of CVs / Resumes advertising for their job roles
In order to minimise their time in sorting out and have a benchmark way of comparing candidates, you’ve been tasked with the challenging task of assisting their problem
Contest Develop a parser that is able to parse through CVs / Resumes in the word (
doc or
docx) / RTF / TXT / PDF / HTML format to extract the necessary information in a predefined JSON format
If the CVs / Resumes contain any social media profile links then the solution should also parse the public social profile web-pages and organize the data in JSON format (e
g
Linkedin public profile, Github, etc
) Solution This Resume parser can run throught unlimited number of Resumes and get relevant info from that
With full-feature installation it supports most of the common use formats, provided by textract : HTML PDF DOC RTF DOCX XLS PPTX DXF PNG JPG GIF application/javascript All text/* mime-types
Pre-Requirements Current solution tested on Windows 7 x64 Maximum (with babun shell ), but it also may run on OSX, Linux
Application is hard dependend on text extracting library textract
Fast install Project is nodejs cli application with some dependencies
If you already have installed copy of nodejs, you can just clone this repo and run npm install : git clone git@github
com:likerRr/code4goal-resume-parser
git
npm install Step-by-step fresh installation First, go to nodejs site, download and setup it for you platform Then, clone this repo git clone git@github
com:likerRr/code4goal-resume-parser
git Run npm install in terminal from root folder of project to setup dependencies At this moment application will work fine, but! By default it supports only
TXT and HTML text formats
For better performance you should install at least support of
PDF (and DOC )
Here is instructions, how to do it from textract README file: PDF extraction requires pdftotext be installed, link DOC extraction requires catdoc be installed, link , unless on OSX in which case textutil (installed by default) is used
DOCX extraction requires unzip be available (e
g
sudo apt-get install unzip for Ubuntu) Please, note, that it’s not necessary install support of all formats but preferably
As for me, I didn’t get setup catdoc for
DOC files under Windows 7, so I played only with
TXT , HTML ,
PDF formats, but I know, it will also work with the rest formats 🙂 Run When you finish installation it’s time to run application
Just put some Resume files to /public (it already has 3 for tests) directory and run in terminal node app
js from project’s root
Then you can access JSONed results in /compiled folder (all file there will represent JSON string of parsed data
Execution presents as dialog between HR manager , that has a lot of Resume to work with, and ParseBoy , who volunteered to help with it, i thought that it should have some fun
How it works Base principle on how parser works, based on dictionary of rules of how to handle Resume file
So we have /src/dictionary
js file, where all rules places
It represents javascript object with the following structure: {
titles: {},
profiles: [],
inline: {},
regular: {}
} All of these keys titles , profiles , inline , regular are converted to regular expressions, that handled by specific conditions: titles – fires on each row of file
If string matches title, so it will capture all text between current title and next title except current
For example we have such dictionary file: {
titles: {
// values are the signs of the key that possibly may appears in the Resume
objective: [‘objective’, ‘objectives’],
summary: [‘summary’],
}
} And next Resume text is: OBJECTIVE Seeking a challenging position to use my software Web development and process optimization skills
SUMMARY I worked on a wide range of products including building advanced dynamic multi language web sites, internal and external API’s, well as creating new internal workflows
If we now run application it will go through next Application Loop (AL): Remove unnecessary Resume file from any \n\r\t and trim all lines Compile rules to regular expressions Split file into lines, delimited by \n Check each line for a match for each title rules When match found, parse text between current title and next title into titles or until EOF Save parsed text (if found) under title key ( objective or (and) summary ) So, according to this loop in the end we will have following JSON file: {
objective: ‘Seeking a challenging position to use my software Web development and process optimization skills
‘
summary: ‘I worked on a wide range of products including building advanced dynamic multi language web sites, internal and external API’s, well as creating new internal workflows
‘
} profiles – fires on each row of file
If profile rule represent an array, so first key will be the name of key and second key will be an handler
If profile rule just a string, parser will try to found matched url without parsing it
Example: profiles: [
[‘github
com’, function(url, Resume, profilesWatcher) {
download(url, function(data, err) {
if (data) {
var $ = cheerio
load(data),
fullName = $(‘
vcard-fullname’)
text(),
location = $(‘
octicon-location’)parent()
text(),
mail = $(‘
octicon-mail’)parent()
text(),
link = $(‘
octicon-link’)parent()
text(),
clock = $(‘
octicon-clock’)parent()
text(),
company = $(‘
octicon-organization’)parent()text(); Resume
addObject(‘github’, {
name: fullName,
location: location,
email: mail,
link: link,
joined: clock,
company: company
});
} else {
return console
log(err);
}
//profilesInProgress–;
profilesWatcher
inProgress–;
});
}],
‘stackoverflow
com’ ], It looks quite a big, but very flexible
So here we can see, that profiles contains two rules: github
com with callback and stackoverflow
com
When profile rule enters Application Loop (AL) and it has valid callback, so it will try to request profile page from Internet and parse data on requested page, according to rules in callback
Then it places all data into Resume object under the represented key ( github in out case)
If rule is just a string and it meets match in AL row, so it simple puts profile link to profile key in Resume object
inline – fires on each row of file
It converts to regular expression, that matches all data after that: expr+”:?[\\s]*(
*)” Example: inline: {
skype: ‘skype’
}, Text: skype: sweet-liker Result will be skype key with sweet-liker value in Resume object
So it can be extended with simple lines of data, e
g
address or first name or whatever
Note, that these rules are unreliable, cause can touch sensitive data from context, e
g
“I don’t have a skype, but I have IM”
After parsing that string data in Resume will be as key skype and value but I have IM
So use on your own risk
regular – fires on full data of file
It just search the first matches by regular expression, e
g: regular: {
name: [
/([A-Z][a-z] )(\s[A-Z][a-z] )/
],
email: [
/([a-z0-9_
-]+)@([\da-z-]+)([a-z
]{2,6})/
],
phone: [
/((?:+?\d{1,3}[\s-])?(?\d{2,3})?[\s
-]?\d{3}[\s
-]\d{4,5})/
]
} Will try find name , email , phone by expression sign
Generic format This solution hasn’t generic output format of JSON string, cause it filled if rule in dictionary match the condition
So, the full possible data, that may be extracted from Resume may have such format: {
objective: ”,
summary: ”,
technology: ”,
experience: ”,
education: ”,
skills: ”,
languages: ”,
cources: ”,
projects: ”,
links: ”,
contacts: ”,
positions: ”,
profiles: ”,
awards: ”,
honors: ”,
additional: ”,
certification: ”,
interests: ”,
github: {
name: ”,
location: ”,
email: ”,
link: ”,
joined: ”,
company: ”
},
linkedin: {
summary: ”,
name: ”,
positions: [],
languages: [],
skills: [],
educations: [],
volunteering: [],
volunteeringOpportunities: []
},
skype: ”,
name: ”,
email: ”,
phone: ”
} Extending All ‘action’ are by building dictionary
js file
For now it has only basics rules, that I met while develop this solution, but it’s very flexible (although a bit complicated) and extensible
Just put your rule according to existing and following main principles and enjoy! Vocabulary Resume object is a place, where all parsed data saves
After parsing whole document it will stringify to JSON and save on into /compile folder
AL – Application Loop: Remove unnecessary Resume file from any \n\r\t and trim all lines Compile rules to regular expressions (under hood) Split file into lines, delimited by \n Check each line for a match for each title rules When match found, parse text between current title and next title into titles or until EOF Save parsed text (if found) under title key ( objective or (and) summary ) Technologies / References Application built on javascript with nodejs 0
10
31 under Windows 7 x64
This application on github Dependencies are: cheerio colors mime request textract underscore In action