EXTRACTION ACCURACY QA

JOB ADVTS. DATABASE (JAWS)

BLOCKS

26 Nov 2003

Sanjeev

BLOCKS

·     Enclosed find some UI’s which we developed in the early-stages of Recruitguru. Then, for some or other reason, we dropped these in favour of what we use currently.

·      Two important UI’s you should know about is

-    “Extraction of entire BLOCKS

-    Breaking up of keywords into categories.

Fields to be extracted in Resume Extractor

1.  Name

2.  D.O.B.

3.  Gender

4.  Current Company

5.  Actual Designation

6.  Total Experience

7.  Education Level

8.  Address

9.  City

10.  Country

11.  Pin code

12.  Home Phone

13.  Mobile

14.  Fax

15.  Fax

16.  E-mail Id

Blocks identified in Resume Extractor

1.  Educational

2.  Objective

3.  Experience

4.  Skills

5.  Personal Details

6.  References

TROUT'S MANTRA : DIFFERENTIATION

17 Nov 2003

Raju/Sanju/Kartavya,

   

TROUT'S MANTRA : DIFFERENTIATION

"Before you break the rules, know the rules,"

First rule, attaining success to be different.

"Effective strategy is all about differentiation,"

brands needed to create a real reason to buy and not a meaningless slogan.

to ensure differentiation by attributes, which was the only characteristic that would make products unique.

How each of them were able to stand for attributes the ultimate driving experience, safety, engineering and styling.

Consumers want to believe that products can contain a magic ingredient that will improve performance.

Where most consumers did not bother to know what Trinitron was all about.

Do not ignore the competition, focus and differentiation are critical in competitive world and CEOs must be willing to encourage sacrifice instead of growth. 

If,

BMW              = Ultimate Driving Experience

Volvo              = Safety

Mercedes         = Engineering

Jaguar            = styling

Then,

Guru Mine        =

Guru search     = Short listing competence not resumes?

Recruitguru      = The future of webservices

3P jobs.com     = For Executive search, corporate India's First choice

Remember

Owen corning (India) Ltd, has even managed to get a trade - mark (not patent) on color "PINK"!

HEMEN PAREKH

NEW MODULES

GURU SEARCH

GURU MINE

COMMON FEATURES

OVER VIEW

INBOX READER

LOGIC

DESIGN OF STAND OUT

MATCH INDEX KILL THE JOB SEARCH

MASTER LIST OF KEYWORDS = 100 / Contained in ERP Function | Keywords = 1,000

Keywords in Pooja’s Resume	ERP Function: Finance, Analytics, Developer                                                 Date: 10-11-03
	Oracle	Cummins	Wipro	Info	Sterlite		XYZ	ABC	LMN
A	Ö	Ö		Ö
B		Ö		Ö
C	Ö	Ö	Ö	Ö	Ö
D				Ö
E	Ö		Ö	Ö
F			Ö	Ö	Ö
G	Ö		Ö	Ö
	UNDERLYING PREMISE / ASSUMPTION • Any given resume (Candidate) would match a large no of job advt (jobs) int to verify degrees. Job-descriptions/ Man-specifications/ skills requirements etc. mentioned in job-advts. Are broad enough to match/attract a whole lot of candidates, each of whom finds, some “degree  of match” between the advertised jobs and his own skills/acknowledge/experience/qualification etc. etc.   • This implies that, whereas a candidate may consider a given job-advt to be an IDEAL job for him (100% match between what he possesses & what that job advt. presents / requires) there will be some other jobs, job advts which he thinks are an excellent match (80%) or Good match 80% or Good match (60%) or a FAIR match 40% or a POOR match 20%   Questions: 1) So simply by looking/reading a job-ad (a whole lot of these, really), how exactly does the candidate’s reachsuchcondition abot 100%-80%-60%-40%-20% MATCHES? 2) what prescisiely is the process (apparently intuitive) that his brain employs, to arrive at such Conclusuions? 3) can we design / develop a self – learning ? Software, that can MIMIC this process & reach draw (mathematical / Statistical) Conclusions?
80			4/80	7/80	2/80	20/80	60/80	400/800	35/80
0.6	.003					.02	.52	.449	.30

 

ANSWERS

➤ Candidate (human) brain COMPARES the keywords contained in any given job-ads  with his own skills/knowledge/Exp/goals (*which two are really keywords contained in his resume!). Of course, he does not do it consciously.

    → make a list of keywords in a given job-ad

 

and then match/compare, as to how many of these (Du-sets) match. But he does this at a sub-conscious level + forms some sort of global/overall impression.

This PERCENTAGE-MATCH.

➤ Equally sub-consciously, he (his brain) assigns "WEIGHTAGES" to different keywords (in the job-ads).

→ In an advt for MATERIAL MANAGER’s position, the candidate’s brain tells him that the keyword Supply Chain Management” carries a higher weightage than the keyword “Discounted Cash Flow!” So his brain KNOWS (possibly thru reading of hundreds of job-ads for the position of MATERIAL MANAGER!) that what are “major” keywords for this position, and what are less important keywords. And obviously, from these FREQUENCY of USAGE/OCCURRENCE of keywords, over thousands of positions, vacancies, the human brain has formed RULES (obviously undocumented!) about the relative-importance (more than/less than) of thousands of keywords in the context of hundreds of jobs/positions!

And one such rule will be:

➤ The keyword “Discounted Cash Flow” carries a higher weightage for the position (job) of a FINANCE MANAGER, as compared to the keyword “Supply Chain Management.”

So the rule reverses itself, based on the “Job/Position” being advertised!

➤ So the weightage (importance) that a human brain assigns to any given keyword is NOT absolute/standalone/independent.

The weightage is *dependent/relative*!

So the same keyword would have different weightage when used in relation to different JOBS/POSITIONS (i.e., Vacancy-Names).

Not only that.

Even within a given FUNCTION (e.g., Material Management), the same word (e.g., Supply Chain Management), would have different weightages depending upon hierarchy/designation-level.

 "Supply Chain Management" would have, let us say, 0.2 weightage for the position of **Stores Officer**.

- A weightage of 0.4 for position of **Purchase Engineer**

 - A weightage of 0.6 Purchase Manager

 - A weightage of 0.8 Materials Manager etc.

➤ So, we have a following MODEL

 

                  (Graph illustration)

    - Keywords (10,000?)

    - Function (5.0?)

    - Basic Levels (10) (or hierarchy levels/actual designations)

    - (2000?)

➤ The total Combinations (10000 × 2000 × 50 = 10,000,000,000 (1 Billion)) are staggering!

➤ But human brain is a *marvellous computer*! It has some very *brilliant* "Approximation Algorithms" which cut out these clutter & quickly arrive at some *BROAD conclusions*. Human brain is also having a billion neurons and is a *parallel-processing machine*.

➤ Till we can afford such complex/huge Computers and equally complex   software we must make do with our existing *P4 machines* or simple statistical software packages.

➤ Fortunately, for our Function Profile Graphs (for resumes) we have already selected "keywords" & also computed their "weightages".

➤ In Phase I we could:

  ➝ Distribute/divide/segregate thousands of Job-Advt. **Function-wise** (subcategories)

  ➝ Using above-mentioned keywords & weightages, compute the *RAW-SCORE* of

     each Job-Advt within a given **FUNCTION**.

  ➝ Plot:

 Now, let's take Resume for Mr. Hidhe (also belonging to Mat-Mgmt function) & find out what is his *RAW-SCORE*.

Let us say it is **40**.

Hence, Matches (Resume) **40** Raw Score is **100% match** with all those *Job-Advt.* which have

*Raw-Score* = **40**.

And we know there are **60** of them.

But Mr. Hidhe's resume with raw-score of **40** is only **50% match** with (maybe) 10 job-adverts whose raw-scores are **80**.

 

By repeating this process we can come up with a frequency distribution graph as shown on *pt.1*, where we can say:

If we want Mahatre’s resume to match with Job-advts.	Then no of job advts available in our database having such Match percentage	Becase
100%	60	All 60 job advts have raw score of 40, (what Mahatre’s resume got )
50%	10	All 10 Job advts have raw score of score of 80

 

Please remember that in **Phase I** we are taking into consideration, just ONE criteria of **FUNCTION**, and comparing **RAW-SCORES**, scored by RESUMES and raw-scores scored by **JOB-ADVTs** both belonging to the same **FUNCTION**.

We are totally ignoring:

➤ Designation (Actual) or Designation Levels in order to simplify calculations/plotting.

This is a good beginning. In **Phase II** we will further refine by adding the dimension of **Designation Level** & work-out Keyword weightages for each UNIQUE combination of **Function AND Design-Level**.

But even in **Phase I**, we will be offering a **SCIENTIFIC / SYSTEMATIC / LOGIC-BASED / STATISTICALLY VALID** matchmaking.

**Match-Dreams**

➤ This is a *GREAT* convenience for a Job-Seeker who is on the job-site often today.

➤ One-shot & He knows *not just weightages* but also *availability* for **you the moment you post your resume!** Detect the different/latest **KILL JOB-SEARCH**!