Naughty or Nice? Santa & Sentiment Analysis

  3 m, 42 s

Every Christmas Eve, I visit more than 378,000,000 children around the world to deliver their presents. For years, no amount of eggnog could cure that logistical headache. My elves were overworked and my reindeer were exhausted. Even my sleigh stopped jingling (rusty bells, you see). But then I discovered sentiment analysis, and my bells started ringing again.

Naughty or Nice? The Challenges of List-Building

In the North Pole, optimization is everything. Even my elves have KPIs to meet! It’s critical for us to know who will get coal and who will get presents as early as possible.

To optimize the time I spend at each child’s house, I classify their behavior-to-date into two buckets: Naughty and Nice.

I base my behavior analysis on language data, which I find to be the most straight-forward and exhaustive source. For example, a child who uses mean, insulting words and phrases is far more likely to be demonstrating Naughty behavior as they speak.

Mind you, I account for both the child’s language data and the language data of their parents and guardians! When Natasha is scolded by her aunt, that intensifies the weight of naughtiness in her overall behavior score.

Now, parents and guardians aside, consider for a moment that (by my calculation) the average child speaks 3,500,000 words a year. Multiply by 378 million, and we have a lot of natural language data to work from! So, when I’m making my List, and while my team of Data Scientist Elves are checking it twice, we must emphasize both precision and recall.

Now, what classifies as Naughty, and what categorizes as Nice? Each category unfolds into their own taxonomies. The details are, of course, pro-ho-ho-ho-prietary.

But in the Christmas spirit of giving, here are a few examples:

Naughty: stupid, dumb(-dumb), mean(ie), rude

Nice: kind, smart, love, well-behaved, sweet

For years, my elves had to take time off from making toys to build and maintain these Naughty or Nice taxonomies, and the List itself, by hand. And when parameters went out of whack, we’d have to take our attention away from toy-making. This led to some big Christmas toy shortages!

The Great North Pole Rocking Horse Crisis

One such disaster happened in 2012 when we discovered an error in one of our configurations that threw the integrity of the whole List into question.

As it turned out, we hadn’t thought to adjust our taxonomies to account for changing language in the era of video games. The phrase “That’s sick” used to always be negative. But not anymore!

It took a whole cohort of elves an entire day to tune our system to account for this one phrase. Meanwhile, farther down the production line, we’d already reduced rocking horse production by 11%!

Between you and me, a couple of kids in Massachusetts didn’t get the rocking horses they were supposed to.

Nuanced Naughty or Nice Behavior Analytics

One year, one of my elves stumbled across Lexalytics. It was like we were the ones receiving a Christmas present! Now we can quickly process massive sets of tricky language/behavior data and then use automated sentiment analysis to build our Naughty or Nice Lists.

Artist’s rendering of Santa’s Semantria Storage & Visualization dashboard

As we acquire more data, my Data Scientist Elves update the Lists and even perform some tricky historical/predictive analytics based on trends and patterns over time.

Meanwhile, their their language packs mean we never need to deal with sloppy translations. Their custom machine learning services team built us a model to correctly classify “sick” as positive or negative. And the support they offer is second to none. If something goes wrong the day before Christmas Eve, I know they’ll have my back.

Lexalytics offers just the right combination of customizable technology and reliable, easy-to-use interfaces. Using their platform, my elves and I maintain an intuitive Coal-or-Presents Dashboard based on nuanced Naughty or Nice Behavior Analytics. This data helps us improve our Time to Gift Delivery while reducing our average Gift Delivery Error Rate.

The best part of this optimization? Of course, it’s the joy of hearing each child’s laughter and delight on Christmas morning.

But the second best part? I get to spend more time relaxing with some milk and cookies by a roaring fire!

And on that note, I must to see to the buffing of the sleigh runners.

Happy Christmas to all, and to all a good night!

Santa

Categories: Sentiment Analysis, Special Interest