JOB MINER

[A] How to figure-out, the "INDUSTRY" of the advertising company? Job Advts do not mention this!

This is a critical issue - since all job-seekers are keen on working in 3/4 specific "Industries" of their choice. So they wish to look-up "interesting" job-advts from those "Industries" only - so, figuring-out the advertiser-company's INDUSTRY is a MUST, for purpose of match-making.

This can be done in following ways:

# Link each "Company Name" with one/more "Industry-Names".

We have already developed such linkages for over 20,000 Companies.

DCA \oplus contains names of 6 Lakh companies - along with their "product/service" categories (similar to Industries). There are only a few hundred UNIQUE categories which can, in turn, be linked with our own 52 INDUSTRY-NAMES (may be Inder Sethi, already did this!).

The only hitch is that many job-advts are released by Placement Agencies - who, do not reveal the names of their Client-companies!

In respect of resumes, such a Software Tool was developed Some 8/10 months ago. Using this tool, Mr. Anjaria had "categorised" some 1000+ resumes, as belonging to different Industries. In the tool, there was a drop-list of 52 industries. After reading the resume, Anjaria concluded that the Resume belonged to a particular "Industry" - which he clicked. Then, below the drop-box, there was another box, which said,

"Why I think this resume belongs to this Industry - which keywords in the resume lead me to conclude this?"

Of course, processing of 1000 resumes, may not have resulted in a sufficiently large enough knowledge-base! May be we need a human expert to process 10,000 resumes like this before we manage to build a respectable Knowledge-Base for 52 different industries.

However, there are \mathbf{2} problems:

(1) Can we safely assume that

\left\{\begin{array}{c}\text { Industry-specific } { KEYWORDS } { of a Resume }\end{array}\right\}=\left\{\begin{array}{c}\text { Industry-specific } { KEYWORDS } { of a Job-Advt }\end{array}\right\} ?

If we have slightest doubt, this is not a big issue. In the very same tool which Anjaria used, instead of loading/processing "Resumes" we could, simply load/process "Job Advts"! The process remains Same - only the document being processed changes! This is fairly simple.

For processing 1000 resumes, Anjaria took nearly a month! He could barely read/interpret & highlight some 30/40 resumes per day.

Capturing human knowledge and embedding it in a software is a slow process! To some extent, this process can be speeded-up, if, instead of loading the full/entire JOB ADVT into the tool, we were to Index (thru IS4S) the "Words" contained in each job-advt and then Only load these words into the tool. Then perhaps, a human-expert can categorise 200 job advts in a day, thru highlighting of Keywords! So, one expert, can, perhaps assign INDUSTRY-NAME to, may be 5000 job-advts in a month (- along with its highlighted Keywords).

A still (much) faster method of developing KNOWLEDGE-BASE OF INDUSTRY-SPECIFIC KEYWORDS,

So, it is obvious that, when a Company posts its job-advt on any of these sites, it has to select a given INDUSTRY-NAME (possibly from a dropdown list).

This means, all Job-advts posted on these \mathbf{3} sites are tagged with an Industry-Name. And this tagging is done by a human expert! (- the recruitment manager who filled-up the job-advt form online).

When human-experts have already categorized thousands of job-advts into different INDUSTRY-CATEGORIES, then why re-invent the wheel?

We have already downloaded thousands of job-advts from these jobsites. We can further download many more thousands. Probably, we already have more than 100,000 job-advts from these \mathbf{3} jobsites - each one duly allotted an Industry-Name by a human expert.

All we have to do is to "group"

NOW, for each INDUSTRY-NAME-GROUP, create a separate folder & then Index thru IS4S. Then arrange all the words found, in descending order of occurrence, and compute \rightarrow their "probabilities" of occurrence.

Perhaps, in each INDUSTRY-NAME-GROUP, the top 50 keywords, would add-up to, may be 90\% of all occurrences.

PRESTO!

What a human-expert, using the highlighting tool described earlier, would have taken months to do, the SORT/INDEX/EXTRACT/ARRANGE/COMPUTE PROBABILITY would take one/two days at most!

Remember, several hundred human-experts (HR Mgts) have embedded/captured their knowledge in \mathbf{1} LAKH job-advts, while drafting (composing the same & then allotting INDUSTRY-NAME to each.

Why re-invent the wheel?

Therefore, in order to do an accurate job, we need to separate-out the "Industry keywords" from "Function keywords".

And this separation must be carried out BEFORE we calculate their "occurrence-frequency" - and the probabilities.

Reason is obvious.

The "Function"-related keywords are "polluting" the Industry-related keywords!

What we need is a population of "pure" industry-related keywords only. And then, we need such populations, Industry-wise.

One population for each "Industry".

To do this separation, we would need

\text{SEPERATOR OF KEYWORDS}

Advt. Source:	Advt. No: