Why one data scientist thinks Facebook is adding a dislike button

A "like" on Facebook is not a strong enough signal for purchase intent, explains Certona CTO Geoffrey Hueter

facebook like button 000021488766

There has been a lot of speculation why Facebook is adding a dislike button to its repertoire. Yes, I know…CEO Mark Zuckerberg said it wouldn’t be a dislike button — that would be too negative — but rather an icon for users to signify that they have read and acknowledged a certain post concerning things like the announcement of someone's death, but didn't "like" it.

Which is all well and good, but chances are no matter how it appears to Facebook users — perhaps this icon will be an emoticon or emoticons of various feelings — it will be acting as a dislike button behind the scenes.

That at least is a theory offered by Geoffrey Hueter, CTO and co-founder of Certona, a provider of real-time personalization technology. Hueter, though, is not a marketer, or rather he doesn’t have a marketing background.

Rather, he has a Ph.D. in physics from the University of California at San Diego,  where he was part of the team that developed the Gamma Ray Observatory — and where he studied neural networks with industry pioneers Robert Hecht-Nielsen and Bart Kosko

His CV also includes stints as a staff scientist managing the development and commercialization of an intelligent machine vision system and manager of several Department of Defense Small Business Innovation Research programs for neural network control systems.

So when he says he has a pretty good idea why Facebook is heading in this direction, I am inclined to listen.

In a nutshell, it is this: Facebook needs a better signal about people's preferences. Its like button draws in plenty of information about what people say — and "say" is a key word here — that they want or appreciate or "like." However, that data, as rich and robust as it is, is not enough for targeting, especially if Facebook hopes to make inroads into social commerce, which so far it hasn't. Ads and messaging — oh, heck yes. But not e-commerce. At least not yet.

Remember the curly fries theory?

It is actually an astounding theory, especially considering the immense amount of data that Facebook has been collecting about users via its like button — not only from Facebook itself, but as a recent blog post by Facebook Chief Privacy Officer Stephen Deadman makes clear, also from sites that have embeddable Facebook buttons.

In a LinkedIn pulse blog post, Christian Rudder, author of Dataclysm: Who We Are (When We Think No One's Looking), and co-founder and former president of the dating site OkCupid, reminds us that, based on data generated from Facebook likes, people can tell how smart we are. One indicator of intelligence — liking curly fries.

Rudder is referring to a study published by University of Cambridge researchers that found that Facebook likes:

..can be used to automatically and accurately predict a range of highly sensitive personal attributes including: sexual orientation, ethnicity, religious and political views, personality traits, intelligence, happiness, use of addictive substances, parental separation, age, and gender.

But all that data is not enough, it appears, and Rudder's post hints at why. As he put it, "...while Facebook does know a lot about you, it’s more like a 'work friend' — for all the time you spend together, there are clear limits to your relationship."

We like everything

The problem for Facebook is that people, at least on social media, like everything, Huerter explains.

"Facebook's like button is not the most reliable signal about a person's intent, especially a person's intent to purchase something," he says. "You also need to know what someone dislikes to make the best offer."

Ratings and reviews are very important signals, he adds, because they provide some of this information.

Indeed, a recent study by the Medill Spiegel Research Center found that shoppers are most likely to buy a product ranked between 4.2 and 4.5 stars in online reviews, but as the rating gets closer to a perfect 5.0, purchase likelihood drops. The researchers concluded this is so because buyers assume five stars are too good to be true.

And so it goes with Facebook users clicking at likes with wild abandon — a problem exacerbated by retailers and brands begging consumers to like them on Facebook and offering a discount or deal if they do.

Twitter is the better signal?

Certona is focusing its research on Twitter as a better sentiment signal, Hueter says.

"You can see how people are talking about the brand on an individual and collective basis to see if that will drive sales," he says. Hueter adds that he is not "totally convinced" Twitter is the better signal but that is where it is doing more research.

The problem with big data

This question of which is the better signal — Facebook or Twitter — for future purchases is a side thing with the company. Its focus, instead, is applying advanced computing techniques to understand online interactions and using those interactions to form a description of the subjects studied. That description can then be used to predict their affinity for an item or belief system or, well, anything.

It uses behavioral models instead of relying on the descriptive data that has been collected on the vast majority of people active on the Web. Or, for that matter, stated preferences of people (such as their "like" data).

From there, the firm develops affinity profiles — or multidimensional clusters as they are called in Certona's world — against which to match a subject. One conclusion the firm reached long ago: stated preferences are usually different from how people actually behave.

Another conclusion it has reached: a taxonomic approach is too limiting.

Let's walk through all this. This, according to Certona, is how it can determine intent online:

It monitors interactions of a predefined group of people, including transactions and communications.

It profiles the subject and objects in those interactions.

It compares the subject profiles to profiles of subjects with known characteristics to determine whether the test subject fits in that group.

It doesn’t rely on expert classification or taxonomy, as noted above.

Hueter explains why this is important:

"When you classify something taxonomically — then you only understand the 'things' for which you define associations," he says.

"A taxonomic approach means that you won't know how a coffee cup relates to a car, for example. But that could be important in the population of people you are studying: they will interact with all kinds of things and express affinity for them, either explicitly or implicitly."

The company recently secured a patent for this process.

Certona is about to announce another patent shortly, Hueter says. It is on a segmentation process the company developed using people's purchase history.

Purchase history, of course, is a very definite signal about future intent. Unlike, as is becoming more and more clear after time spent with Hueter, Facebook's "like" data.


Copyright © 2015 IDG Communications, Inc.

Bing’s AI chatbot came to work for me. I had to fire it.
Shop Tech Products at Amazon