Using NLP to find out what motivates your employees!

It is something along the lines of making the subject more interesting, method of teaching more engaging, and to align the sequence of contents of a subject with a child’s natural curiosity.

Easier said than done.

Having a fatherhood experience of about 2 months, I don’t feel comfortable advising on that topic.

For that, you might get some useful tips from “Unconditional Parenting”, and “Punished by Rewards”, both by Alfie Kohn.

It is all nice and clear (or at least less confusing) when we think about this argument as it applies to kids, but things get a little complicated when we bring adults into the picture.

Pink goes on to argue that the detrimental effects of rewards, such as “pay for performance”, applies to you and your colleagues as much as it applies to kids.

If you are a hardworking employee waiting on the raise that is long overdue, you are probably like:This is not to say that a company can get away with paying its employees peanuts if they are intrinsically motivated.

It has still got to pay its employees generously, not to motivate them, but to remove them from thinking about it too much.

To be honest, I am as skeptical as you about the effects and practicality of completely weeding out extrinsic motivation in a business setting.

But that doesn’t undermine the usefulness of the book in convincing us to at least practice caution while using extrinsic motivation and take intrinsic motivation more seriously.

Most of us, at some level, knew this all along — inherent interest in work leads to better performance and happiness than working just for the sake of recognition or money.

In fact, we have known this for at least two millennia.

Therefore, I pray you, my dearest Lucilius, do the one thing that can render you really happy; cast aside and trample under foot all the things that glitter outwardly and are held out to you by another or as obtainable from another!-Seneca (Letters from a Stoic)On action alone be thy interest, never on its fruits.

Let not the fruits of action be thy motive.

-Bhagavad GitaSo what is the use of these books that reiterate the knowledge that is already buried deep in us? — To split this knowledge into components, and capture the essence of these components in precise words.

Once we have these feelings captured in words, we can play, learn and gain more from them, and that’s exactly what we are going to do.

The book theorizes that intrinsic motivation has one or more of the following attributes:Mastery —Our urge to get better at our craft, and work on challenging projects.

Autonomy — Our desire to be self-directed.

Purpose — Our desire to do something that has meaning and is important.

We are going to look into a code that labels the glassdoor reviews that showcases these qualities as “intrinsically motivated” and essentially gage the health of intrinsic motivation of Software, BI and Engineering department of few companies.

I used Matthew Chatham’s awesome scraper to scrape about 20,000 employee reviews of Amazon, Google and DXC technologies from Glassdoor.

Our focus will be on the overall rating, the job title of the reviewer, and the Pros.

Within pros, our model will identify the green highlights as intrinsic motivators, and the yellow highlights as extrinsic motivators.

This is done bycomputing gensim’s n_similarity(cosine similarity) of these words with the ones in the corresponding lists — intrinsic_motivators=[‘interesting’, ‘challenging’, ‘exciting’, ‘learn’, ‘autonomy’, etc.

] and extrinsic_motivators=[‘pay’, ‘benefits’, ‘perks’, etc.

] using vector representations in Google News word2vec.

labeling the review as “intrinsic” and/or “extrinsic” if the n_similarity score with the words in the corresponding lists is above a specific threshold -0.

6 (chosen by trial and error)To reduce the number of n_similarity computations, we will consider only the nouns, verbs, adjectives, and adverbs (extracted from part_of_speech tags — spaCy).

Example: [([‘great’], ‘ADJ’), ([‘culture’], ‘NOUN’), ([‘’], ‘PUNCT’), ([‘lots’], ‘NOUN’), ([‘of’], ‘ADP’), ([‘opportunities’], ‘NOUN’), ([‘to’], ‘PART’), ([‘move’], ‘VERB’), ([‘around’], ‘ADV’), ([‘’], ‘PUNCT’), ([‘good’], ‘ADJ’), ([‘benefits’], ‘NOUN’)]The department of each reviewer is guessed by checking if the words in the title (‘Software development engineer’) belong to a list of words related to that department (SW_Eng=[‘sde’, ‘web’, ‘software’, ‘engineer’, ‘developer’,etc.

]).

Note that the plot above does NOT represent exclusive intrinsic motivation.

The reviews classified as “mentions intrinsic motivators”, may or may not have elements of extrinsic motivation.

Here is the confusion matrix plotted from about 140 Amazonian reviews that I manually labeled as “Intrinsic” and/or “Extrinsic”.

Some limitations of this model:Misses picking the intrinsic or extrinsic motivators when none of the individual words in a particular sentence lean towards either category but convey the meaning only when put together.

Example: “you are pushed to your limits”.

Missing words such as “rsu”(restricted stock units) and “401K”, which are not in google news word2vec.

This can be avoided by checking if these strings are in the review, and computing n_similarity only if they are not.

Both these limitations are not too big of an issue for our purposes, as they are either rare or most likely to appear with other words in the review that our model can pick.

Example: “401K and stock options are great”.

We will miss “401K”, but will pick “stock”.

The primary objective of this project — to gauge a department’s health of intrinsic motivation, ends here.

What follows is just a nudge to take intrinsic motivation more seriously.

Comparison of ratings by Amazonians who mentioned intrinsic motivators Vs who did not (SW, BI, and Eng Dept):Why compare ratings?.Based on not so wild assumption that employees who give higher rating are likely to be happier (at least at work) and productive compared to the ones who don’t.

Given that the number of employees who mentioned intrinsic motivators is about 80%, absolute counts of two categories in each rating bucket will not tell us anything, as each bucket will obviously have more intrinsically motivated employees.

So, to level the field, we will look at what percentage of employees in each category (intrinsic, no-intrinsic) fall into each rating bucket (1 thru 5).

I know, correlation doesn’t equal causation.

In addition to that, this category is not even exclusive of extrinsic motivators.

So intrinsic motivation might be taking undue credit.

Let’s give extrinsic motivation the same advantage and see how it fares!Too good to be true, isn’t it?.I am with you!.The same analysis on Google and DXC reviews yielded kind of similar results but not as perfect as that of Amazon’s.

Even then, arriving at conclusions from these results wouldn’t be right nor useful.

While we can’t be certain about the effects of nature of motivation on the ratings without diving much deeper and analyzing reviews of a few more companies, that shouldn’t stop us from taking intrinsic motivation more seriously.

Thanks for reading!.Cheers!You can find the GitHub repo here.

References:[1]Matthew Chatham’s Glassdoor review scraper.

[2]Peter Min’s “Aspect-based opinion mining”.

[3] “Drive” by Daniel.

H.

Pink.

[4] Alfie Kohn’s website.

.. More details

Leave a Reply