Three insights from social data

Social data is powerful, goes the saying. But how can abstract data, tweets, likes and repins, tell you anything about an individual, a group or the world?

I have heard from marketers that they doubt the value of social data. Although Apple’s acquisition to Topsy should tell us something different. Apple sees massive value in social data.

And McKinsey’s robust assessment that social technologies combined with big data could release around $1trn in value across consumer goods, financial services, manufacturing and professional services alone.

There are three classes of insight one can derive from social data.

1. Least valuable are 'on-social' insights (how many likes someone has, or might get);
2. More valuable are inferring otherwise expensive to discover traits (like whether a person or group of people are early adopters, or influential in referring products);
3. Most useful are predictions of states we can't otherwise measure (like predicting the success of given meme, product or movie; or forecasting future stock price movements.)

Through techniques of statistics, in its big data form, machine learning, researchers are developing models to predict many non-trivial features in the last two buckets of insights. Helping us unearth expensive-to-find traits about audiences; or to make predictions about future events.

The PeerIndex science team came up with the following examples:

Predicting influence and strength of relationships:

Predicting who influences who on social media has several potential application areas where it can create huge value - specifically in marketing channel optimisation. Which people should marketers influence (and how) in order to maximise the onward spread of their message?

In this important study Bond et al. investigated the role social influence plays in political mobilisation. Their key finding: “The social messages displayed on Facebook walls directly influenced political self-expression, information seeking and real-world voting behaviour of millions of people.”

Combine this with the fact that it is possible to predict your political orientation quite easily, social data will probably play an increasing role in elections and political campaigns in the future.Yes, beyond canvassing for votes and mobilising, MoveOn style, but in the targeting, prediction and optimisation of messaging and campaigns.

PeerIndex investigates product influence on social networks. Dr Ferenc Huszar, who leads our data science efforts, says "Inferring influence has clear value to consumer brands, as it allows them to find people whose referral value is high. Companies have been focussed on Customer Lifetime Value, which is defined as the cumulative profit or revenue from the customer’s purchases in the future. Now they will be able to combine their data with social data to understand Customer Referral Value, a much more powerful indicator that tells you how much value a customer creates by their own purchases and through the value of customers they would refer." Huszar writes about customer referral value on Harvard Business Review.

Predicting trends and adoption rates:
One powerful use of social data is to predict product adoption rates. MoviePilot predicts film success by target market through looking at Facebook likes. This paper demonstrates how one can forecast box-office revenue based on Twitter activity.

A number of companies (including us, PeerIndex, and Dataminer) are looking at detecting emerging trends on twitter before they are trending in mainstream audiences. These type of forecasts, when automated at scale, have tremendous value for brands who want to optimise their marketing or identify new opportunities.

Predicting personality and other traits:University of Pennsylvania researchers recently demonstrated they could predict usersgender and other personality traits by looking at Facebook likes.

IBM is experimenting with targeting ads based on people's personality types. And they aremodelling personality by looking at twitter streams.

For marketers, understanding personality traits of target audiences could be a powerful input into the marketing plan. Understanding softer attributes about customers could shape product development, naming, packing, pricing and marketing communications.

Predicting other events

Wang et all use language models over twitter to predict hit-and-run crimes. DataMinr analyses twitter to find signals that might affect stock prices - before they become traditional news.

Social data is valuable, why don't people think it is?

Strangely, it's about signal-vs-noise. Companies tend to see first order social data, that is likes, followers and retweets. It is hard to prima facie connect these to business outcomes.

The second and third-classes of social insights (the harder to predict but more actionable) are simply rarer. So the market chatter about real insight is drowned out by the trivial, and more commonplace.

However, as smart companies like Apple, start to make bets on the value of social data, you can be certain the others will follow.

I'd be fascinated to hear of other examples, where large firms have used these 2nd and 3rd classes of social data to change business decisions.

What Social Data Can Tell You (And Why)

Pretty much everything. In a nutshell.

Take this fascinating piece of research: Facebook's data scientists were able to determine a very particular pattern when couples were courting, and when they had started a relationship simply. How? Simply by looking at the frequency of status updates between two people.

In essence:

During the 100 days before the relationship starts, we observe a slow but steady increase in the number of timeline posts shared between the future couple. When the relationship starts ("day 0"), posts begin to decrease. We observe a peak of 1.67 posts per day 12 days before the relationship begins, and a lowest point of 1.53 posts per day 85 days into the relationship. Presumably, couples decide to spend more time together, courtship is off, and online interactions give way to more interactions in the physical world.

Social data are powerful

These predictions comes from a new class of data called "social data". Broadly speaking, social data are the data that people create when they use social platforms like Facebook, Pinterest or LinkedIn. It is our likes, pins, favourites, retweets, status messages, content of those messages, people we are friends with.

Social data are normally voluntarily and informally created by the individual during the act of using a social platform. It is voluntarily made public (or semi-public)on the platform in question. It reflects their ordinary course of business, the stuff we care about. It provides a picture which is explicitly incomplete. It can be viewed in aggregate or at an individual (although not necessarily prima facie personally identified).

Social data as a class of data is new. No more than 10 or 15 years old, running back to the older social platforms (although arguably it stretches back to the earliest social protocols like Usenet).

As a class of data it is also very powerful. And can predict much more about individuals and groups that the non-specialist can imagine.

Birds of a feather, that's why

Take the pattern of courtship to relationship discovered by Facebook's data scientists. This pattern emerges because of similarities in the way people behave as they go through similar life stages. We can all be different, but in aggregate we might be able to evince some underlying pattern. It is these underlying patterns which allow us to predict things which seem unlikely - like whether a couple of entering a phase of courtship.

I write 'unlikely' but it's really only 'unlikely' if you don't work with social data, or other behaviour data, and don't hang around with data scientists and other statistical types.

In truth, that we can predict courtship from signals like this is hardly at all surprising. We can predict other things too. These include wealth, gender, sexual orientation, even whether ones parent's divorced before you were an adult. All of those predictions were based on Facebook Like data - commonly acknowledged as quite noisy.

For a moment think about that.

It's a brave new world

For companies, social data allows them to know who they are dealing with, well before that person becomes a customer. If a potential customer follows you on Twitter, you can build a pretty full picture of who they are, what they might like and what you might want them as a customer. Well before you decide to despatch advertising or a sales person towards them. For example, at PeerIndex we've helped companies modify their product offerings in near real-time on the strength of behaviours we've picked up in social data. IBM is helping companies identify customer's personality types based on social data - for better customer service.

For governments, social data allows you to understand the over-arching trends and themes in your jurisdiction. It allows for the early detection of themes that you might need to respond to. (Disease outbreaks? Or sociological changes in household behaviour?)

With the increase in people sharing data with companies - a subject I touched on a few weeks ago - such predictions will become more precise, more accurate and more relevant.