Notes: CLICK-STREAM ANALYSIS

“Freedom of Choice & Recommendation Systems”

(TOI clipping dated 19-07-06 attached)

Pranav → Saurabh → Rahul → HCP

If any of you want to read more about this issue, I can bring from home a good book on the subject titled:
“A Paradox of Choices.”

When (someday) we put into place
Recommendation Systems
→ for Jobseekers (Jobs Recommended)
→ for Employers (Resumes Recommended)

Then, we are venturing into this “territory.”
We are certainly NOT deciding (which job to apply to / which candidate to shortlist) — but we are helping jobseekers and HR managers to quickly “zero-in” on a fewer options (choices), which makes it easy for them to decide.

Google/Yahoo fall miserably on this!

“Buying What Others Have Bought”

Rajeev – Reji – Abhi
16-08-05

Earlier I have sent you a note about showing a counter in job search. I suggest a similar counter in the resume search too.

In short display, these will look as follows:

JOB-SEARCH

Job/Advt Details	How many have looked at this ad before?	While conducting job search	Having received as SMS job alert	Partner websites

RESUME-SEARCH

Resume Details	How many recruitment managers have searched this resume so far?	While conducting resume search	Having received as email job alert	Partner websites

I will explain in person WHY such display is absolutely MUST.

“Buying What Others Have Bought” (contd.)

As far as Job-Search Display is concerned, I have explained the importance of Counter-Window in my note (Aug 16).
What others have chosen?

But this concept is even far more important when it comes to resumes / resume-search.

With the economy booming, there is a mad scramble to hire good candidates and hire them before your competitor makes them a better offer!

But how can an HR manager conclude, simply by looking at the Resume-Search Display, that

– “XYZ seems like a GOOD candidate.”
– “ABC looks like an AVERAGE candidate.”
– “LMJ is a POOR candidate.”

To reach such a conclusion, he would need to
▸ first download that resume
▸ next read it …

But when you are presented with hundreds of resumes in a short display, which one to open / read and which one to skip?

If you open each (one-by-one) to read, you may end up wasting a lot of your time.

The huge number of choices/options presented could
▸ empty you
▸ confuse you
▸ paralyze you

You may even “give up” after opening and going through 20–30 resumes.

And once you “give up” and stop opening further resumes, guilty-consciousness takes over!

Now you feel guilty that you did not open ALL resumes and possibly the better/best candidate you missed out!
You blame yourself for not having opened 640 resumes that were listed in short display — even though you know fully well that it was impractical to open & read 640 resumes in order to find/locate 36 which had higher probability of being “Good.”

This is where the counters shown on-display come in handy.

“Seen” / “downloaded” / “received” by a large no. of HR mgrs, then I, as another HR mgr, can reasonably conclude that it must be a good — they also would be many HR mgrs who took the trouble to download & read it?

To me, as an HR mgr, this is the clear proof that THIS is a GOOD resume.

The large “Number” inside the WINDOW speaks for itself!

So, now I am on safe ground. If I download / open / read only those resumes against which I see a LARGE number, I am selecting a quite sound / rational one. I don’t need to worry for those resumes against which there is a SMALL number.

Another angle:

If a resume has figured again & again, has been delivered again & again, as RAM / RAS,
then it is a CLEAR proof that a large no. of HR managers are looking for exactly such a person!

(Note: “Multiple Consignments for it confirm requirement” written in margin)

Obviously this particular resume was found to be a perfect match against Resume Preference forms filled up by a large no. of HR mgrs.

Obviously this resume met the requirement criteria specified by a large no. of the mgrs.

A guy like this seems to be in GREAT DEMAND!

And this LARGE number appearing in the window is somewhat of POPULARITY INDEX!
(like Indian Idol!)

He has scored MAX. VOTES from peers — professional HR mgrs.

So, I am on a perfectly safe ground, if I too VOTE for him (mentally).

Naturally, so many HR mgrs could not be wrong in their judgement — their collective judgement expressed by a large no.

With this kind of thinking, I am pretty sure that, the more a —

Resume gets clicked / downloaded / opened,
the more it is LIKELY to get clicked!

It can become a self-fulfilling prophecy.

But how can we get this process into motion?

To begin with, the probability that any particular resume gets short-listed during a resume search (and gets opened / clicked) is quite lower — especially if there are a large no. of resumes of that particular skill / exp / background / Ind / function etc.

Again, if the resume was posted only 2 days back, the counter may show ZERO.

If it was posted 2 months back, it may show 19.

And if it was posted ONE YEAR back, it may show 28.

What I mean is,
The probability of a Large No., solely due to Resume seen (count), is low.
It can only increase with time.

But,
this “no.” under RAM / RAS counter could be large (or at least respectable) within 2/3 days of resume getting posted!

If the resume belongs to a person who is in reasonable DEMAND, then it is quite possible that a no. of HR mgrs have submitted similar

“Resume Preferences”
(for getting RAM-RAS)
on different partner websites.

In which case, this fellow’s Resume Alert will get delivered to all these HR mgrs within first 2/3 days.
And will continue to get delivered as each new
“Resume Preference” form gets filled & submitted on websites.

So, even if HR mgrs are not conducting resume search online and not short-listing this resume, they are still getting a “resume alert” under RAM / RAS, and the counter goes up each time.

Then someday, we will find a method to convey to each candidate his
POPULARITY INDEX.

Abhi → Rajeev → Reiji
Aug 10 / 2005

What others have chosen?

In book “Paradox of Choice: Why More is Less,” author Barry Schwartz, writes (p.72, 2005):

“… when faced with a choice among hundreds or thousands of possibilities, the search for something good enough can be enormously simplified by knowing what others have chosen.”

Maybe this is the reason why some job sites indicate
No. of resumes received against this job ad.

I think JobsAhead or JobStreet has this feature – but for Corporate Advertisers to see! Maybe as part of their
RESUME MGMT / TRACKING software.

So that a Corporate Advertiser knows how many resumes were received against each job ad which they posted on JobsAhead / JobStreet.

And by clicking on this NUMBER, they can view all those resumes – along with their “Rating” / their “decisions” – call not call / shortlisted / rejected / etc.
(LEENA – SUVA – TEXSTAR – T. MARK)

I believe you are planning to incorporate some such feature in Global Recruiter.
If our feature turns out to be superior (to JobsAhead / JobStreet’s), then our partner websites will enjoy a definite competitive advantage over them. In India, this is very desirable.

But 3–4 years ago, we had tried to incorporate somewhat similar feature on 3pJobs.com, wherein:

There was a WINDOW ( ) against each & every job-advt posted on 3pJobs.
During job searches, some job-advt(s) would get shortlisted.

Sr	Job Advt No	Advertiser Name	Position	Mgr-Spec	Window
	236				23
	1283				42

Now, if a jobseeker clicked on any of these (to read FULL advt), then the counter in the window would go up — indicating how many jobseekers had “opened” / “viewed” this job-advt.

The rationale / reasoning was simple:

If the “no.” (in the counter window) is large/respectable, the next candidate will be much more likely to click on that advt — out of curiosity & out of belief that

“So many jobseekers can’t be all wrong to have clicked & opened this advt.”

The idea was to create a self-fulfilling prophecy whereby

The more clicked / viewed an advt, the more it will get clicked again!

There would be a “Herd Mentality” effect.

This mentality would set into motion a

VIRTUOUS CIRCLE —

The more the no. of past clicks, the more the no. of future clicks.

This feature can also be incorporated in RESUME SEARCH.

A HR manager is bound to click/open that resume which he sees/knows has been opened earlier many times by his co-professionals!

Let us do this.

Amazon-type "Recommendation System"

For Jobseekers

To recommend to jobseeker, what "search criteria" to select, during his next "job-search".

For Corporates

To recommend to HR mgr, what "search criteria" to select, during his next "resume-search".

This is fairly simple.

Let us first take Jobseeker's case.

We have already decided to enable him to view his own:

Job-Search History
Apply Online History

For conducting a jobsearch, he has to select, "Ind / Func / Desig-level / City".

When he applies online, he is applying against a specific Job Advt.

Once again, while posting a job, a HR manager has got to select — Ind - Func - Desig. level - City".

In anycase, for Job-Search History & Apply Online History, we have to capture/store some fields.

Let us also store fields for "Ind - Func - Desig. level - City" everytime/eachtime, a jobseeker conducts a job-search or applies Online against an advt.

This capturing/aggregating should be ongoing/dynamic process.

After a while, we would get to see a tabulation like follows:

Total No. of Job Searches so far:	54
Total No. of Apply Online so far:	46/100

Ind. selected	Func. selected	Desig. selected	City selected
Name	NO	Name	NO
Pharma	64	Sales	72
Chemical	26	Mktg	16
Packaging	10	Service	12
Total	100		100

This tabulation is strictly for our own viewing (not even by partner websites). In fact, even we would not want to see it!

But, now we know the probabilities with which he is likely to select during his next job-search:

Industry = Pharma $\rightarrow$ 0.64
Function = Sales $\rightarrow$ 0.72
Design. Level = GM $\rightarrow$ 0.68
City = Mumbai $\rightarrow$ 0.58

As soon as he logs in for "Job-Search", the search-box will get CUSTOMIZED & get displayed, as follows:

Dear Ajay, your past search history suggests your preference for Search criteria, shown below. Of course, feel free to modify.

JOB SEARCH

Field	Input Area/Dropdown	Note
Keywords	[Text box]
Min Exp	[Text box]
Industry	[Dropdown] Pharma (64), Chemical (26), Packaging (10)	* Your past preferences, arranged in descending order
Function	[Dropdown] Sales (72), Mktg (16), Service (12)	* Your past preferences, arranged in descending order
Desig. Level	[Dropdown] GM, VP, CEO
City	[Dropdown] Mumbai, Delhi, Chennai
	[Submit button]

For the jobseeker concerned, this will appear amazing! This is "personalization/customization" at best.

Now the jobseeker knows WE CARE, we are here to make his life easy.

This is exactly what Amazon does—and it is NOT rocket-science!

It is a good thing for us that, as of now, no other jobsite does this.

If we do this, it will set us apart, differentiate us. And unless we differentiate, we cannot attract (would-be) partners.

They (would-be partners) need such features to divert Jobseekers & Corporates from Monster/Naukri.

We can, similarly, "Recommend" Ind / Func / Desig. level / cities etc. to a HR manager based on his past "Resume Search History".

In Resume Search U/I, as soon as HR mgr logs-in, we can fill-in / really re-arrange the dropdowns, with most-searched names at the top (the relevant choplists).

This can be outsourced to same person who will develop various ADMIN screens / software / displays.

We can also accumulate/aggregate "Keywords" used by jobseekers during their job-searches & calculate "frequency of usage" & then display those (with high frequency) in their "Keyword Box" when they next log in.

Similarly, when a jobseeker conducts his first ever (very first) jobsearch, as soon as he logs-in, we can fill-in his job-preferences (Ind / Func / Desig. level / City) into the respective jobsearch fields. This is quite easy, and our "recommendation-system" starts from his very first jobsearch!

Here is the text conversion from the uploaded images:

📈 Enhancing the Recommendation System

Keyword Aggregation

Click Stream Analysis and Prediction

Names (User registration flow example, connected to HCL news article context):

Saurabh
Yogesh
Rahul
Swati
Sonal

Click Stream Analysis

Content Analysis = "Click Stream Analysis" to calculate the "frequency" of clicking of each link.
NOT difficult to "predict" if frequency of clicking of each link/choosing of each field-value, is stored in database & probabilities re-calculated everytime.
e.g. Post Job - Cum - Resume Search

A Corporate Subscriber will repeatedly:

a. type, Post "vacancy" / Edit "keywords" / Job Descript
b. click on: Industry , Function , Desig. level , City , etc. etc.

Since most job-advts. (in any company) gets repeated again & again & again (only frequency/interval may vary), it is not difficult to figure out "trends/patterns" — then claim to "Predict"!

Date stamp: 20/12/06

To: Rajeev cc: Vikram cc: Rahul Date: 05/02/06

I have not come across any Indian jobsites which use Click Stream data to analyse and present/display the results for the benefit of jobseekers and employers.

Thru ARM ("clicking of 'Feedback Links'") we have made a small beginning.

But this can be said to be "Tip of the (Analytics) Iceberg".

We can do lot more - as shown in enclosed pages.

Most of this is quite SIMPLE/STRAIGHT FORWARD.

Wherever there is a "droplist", count every click and add-up to aggregate.

Have a Separate Data Table for each Droplist.

Droplists, themselves are first to be separated into:

Droplists in "Jobseeker U/I"
Droplists in "Employer U/I"

Arrange all Data-Tables in descending order of "No. of accumulated clicks".

Such "descending order" will facilitate On-the-fly Generation of:

Bar charts
PIE charts
Distribution Curves This is also Recruitguru's LOGO! Now you see the reason for its selection as our Logo!

If you type in Google's Search bar "Clickstream Analysis Software", you will get 156 results!

Lots of expensive commercial S/W are available with hundreds of features — which we don't need.

Maybe even Google's "Web Analytics" (free S/W) which Athul/Reiji are supposed to have installed on all of our websites, could also do.

What I am suggesting, I don't know.

In any case, Click Stream Analysis to generate lots of Charts, is a LOW PRIORITY.

(Even though I consider this feature/application as VERY IMPORTANT from the viewpoint that it will):

"Differentiate" Global Recruiters from Monster/Naukri/Timesjobs & place us in a class by itself.
Motivate a lot of jobsites to become our partner websites. They can use this feature to "Sell" subscriptions to Corporate employers.
Attract a lot of jobseekers/employers to subscribe/register on partner websites.
Generate for us/our partners a lot of additional revenue, if we "price" each graph/piechart/barchart/frequency diagram VIEWING (click open) at Rs. 5!

Questions

Which no jobseeker or employer has ever bothered to ask, because they know/think that there is no jobsite that can provide "answers" to their questions!

During a jobsearch what "Industries" are jobseekers Clicking on?
Put it other way, what "Industries" are popular/sought-after by jobseekers?
What is the "Popularity Index" of any given "Industry", arrived/derived from CLICK-STREAM DATA of Jobseekers?

Such a question could be easily answered as shown in ANNEX: A.

Similar/identical questions can also be easily answered, in relation to

Which "Functions" are most/least popular
"Desig. Level"
"City"
from "ARCHIVAL METHOD"
Which "Actual Desig" are most/least popular
"Companies"

This is all about "SUPPLY-SIDE"

(jobseekers' "database of intentions")

One can carry-out a similar exercise

with regard to "DEMAND-SIDE"

i.e.

What are the intentions of EMPLOYERS?

(i.e. The Demand-Side).

Questions (that could be answered thru

such Demand Analytics) could be:

Candidates having background (exp)

$\rightarrow$ in ABC industry are in big demand

$\rightarrow$ in XYZ " " " least ".

Maxm jobs are in ABC/HJK cities Minm " " " LMN/OPQ " (see our PIE-CHARTS in WWJ "Where are the jobs?")

Q: What is the "most-desired" Edu. level that employers are asking for? For which Edu. level, there is no demand?

Q: Experience (No. of years) demanded by Employers

U/I = JOB SEARCH / CONVENTIONAL

ANNEX: A

05/02/06

Search Parameter = INDUSTRY

Search Date	Search By (PEN)	Search Serial No.	Adit	Agri	Ariliny	Auto	Banking	Beverage
		1
		2
		3
		4

What are we trying to find out?

During Conventional Jobsearch, what "Industries" are jobseekers clicking on.

What can Employers/Jobseekers conclude from such an analysis?

Which "Industries" are popular/sought-after by jobseekers? $\rightarrow$ kind of "Popularity Index of Ind."

The display can be

Bar charts * Simple Tabulations (in descending order)
Pie chart

There are altogether

10 user Interfaces
10 search-parameters

Of course, not all U/I have all Search - criteria. Nor are all DROP LISTS.

Which means, altogether there could be 70/80 such TABLES, in which, we could compile/aggregate "CLICK-STREAM" data, on a continuous/ongoing basis.

Once-a-day each DATA-TABLE could be Converted into a beautiful/colourful chart/graph with appropriate "TITLE/HEADING & BASIS (PERIOD/NO.S)" (Like "Where Are the Jobs?")

$$8943$$

	23	12	44	612	865	212

Popular Industries may feel confident that they should not have difficulty in finding enough/competent candidates since their Indus is in such great demand. Conversely, least 'popular' Industries may realize that, to attract enough good candidates, they would need to revise their Compensation/perks/benefits/designations etc.

Translate

Wednesday, 19 July 2006

CLICK-STREAM ANALYSIS

No comments:

Post a Comment