Akshay :
You are best placed to decide which
technology to incorporate into tool ( PREDICTOR ), for developing an INDEX of
CULTURAL FIT , between a given candidate and the Client Company
I hope , you are making good progress
and would be ready to deploy PREDICTOR on our server , in 2 weeks
Warm Regards,
Hemen Parekh
+91 98675 50808
Mitchelle :
In one of my earlier mail , I had
suggested an alternative . This consisted of our preparing a list of “ Keywords
/ Phrases “ , against which , a clent can “ tick / rate “ , to tell us :
“ This is our Corporate Culture “
Obviously , this ( filling in / ticking
) need to be done by our client , ONLY ONCE ( - not for each position ) – since
CORPORATE CULTURE does not change from position to position !
Do you think we can get our clients to
do this ? Even by sitting across their table and helping them out ?
This may not be too difficult
What is . however , difficult is to
decide :
“ This keyword ( or phrase ) found
in a candidate’s resume = this keyword ( or phrase ) of a Client’s culture List
“ – and by X %
You may want to discuss with Akshay ,
how best this “ Correspondence / Relation “ can be established
Can software do this from some “ first
principles / rules “ ?
OR,
Can Akshay develop a simple MAPPING TOOL
and give us , where , we can get our Recruiters to carry out the mapping ,
manually ?
In this later case , if our consultants
carry out such mapping for 1,000 resumes , that might be enough starting data
for the algorithm to improve upon later
From: Akshay Sehgal
[mailto:akshay.s@ipredictt.com]
Sent: 05 July 2017 16:22
To: Hemen Parekh
Cc: Rohit Verma; Mitchelle; Nirmit Parekh; Hemant Juvekar; Prashant Sharma; Pradeep Bhalla
Subject: Re: Can NeuralNet Shell help ?
Sent: 05 July 2017 16:22
To: Hemen Parekh
Cc: Rohit Verma; Mitchelle; Nirmit Parekh; Hemant Juvekar; Prashant Sharma; Pradeep Bhalla
Subject: Re: Can NeuralNet Shell help ?
Hi
Hemen,
A neuralnet
based model will always help since they are supposed to be the most
accurate models currently. There is, however, a catch. The amount of time it
takes to build a complex neuralnet to solve a problem (not to mention the
amount of expertise it takes) is mind boggling.
To give you
some context, neural networks were built by hackathon participants to
"calculate the number of sea lions in any given picture". This was
for understanding the population of sea lions and their migration patterns for
environment protection research purposes. It took the best of the minds on this
planet about 1 year to build the model that, if given a picture of sea lions,
can calculate the number of sea lions accurately (with almost 99% accuracy).
Interestingly the winning team comprised of 4 data science PHDs from
China.
Also,
feeding a model the ''corpus'' of 10 stories will not be enough. It needs to
provide the outcomes/predictions along with the corpus as well. So, if I want
to say that a given sentence in the story talks about "Honor", I need
to tag the set of words and sentences to the result "Honor".
There is
a different approach to semantic analysis like you have described in the use
case above. There are massive text libraries on NLTK which comprise of books,
novels, scriptures, presidential speeches etc, where each word/sentence is
tagged by "Part of Speech tagging, POS tagging" (which tags nouns,
verbs, adjectives etc) as well as a thesaurus (meanings of each word). The
combination of these lets us use natural language processing algorithms to
learn from such databases and utilize these learnings to handle fresh documents
of text. This doesn't require CNN but more simply just relies on highly
optimized linear algebra based algorithms to score and match words/sentences on
their meaning.
Thanks,
Akshay.
Co-Founder, iPredictt Data Science Labs
Head, Product Strategy & Research
On Wed, Jul
5, 2017 at 8:51 AM, Hemen Parekh <hcp@recruitguru.com> wrote:
Akshay / Rohit :
In my mail yesterday , I talked about CULTURE FIT between
a candidate and the hiring company
In this connection , I drew your attention to 41 episodes ( stories ) listed at :
Out of these , I had highlighted some 8 / 10 , leaving out 30 +
( altogether , I can send to you approx. 71 stories ;
soft copies )
Here is a thought :
Using my ( human ) intelligence , I tried to find /
locate , in those texts , Words / Sentences , which in my opinion , “ described “ some
values of that organization ( L&T ), and highlighted in RED
Then , at the bottom of each story , I listed “ words
/ phrases “ , describing those VALUES (
culture of L&T ) – in Yellow
Let us treat these 10 stories ( along with those RED
& YELLOW words / sentences ) as some kind of SEED
Can we feed these in a NeuralNet
software and ask it to search / locate and highlight other “ words / sentences
“ from the remaining 30 stories , using forward
propagation ?
If that works out , we could have that MASTER LIST of narratives ( words / sentences ) ,
which describe Corporate Culture ( RED ) !
Thereafter we can develop a simple MAPPING TOOL , where , against each RED , I can enter a YELLOW
!
Is this feasible ?
With Regards,
hemen
parekh
(
M ) +91 - 98,67,55,08,08
No comments:
Post a Comment