Introduction

This book is about using data to understand how social media services are used. Since the advent of Web 2.0, sites and services that give their users the power to actively change and contribute to the services' content have exploded in popularity. Social media finds its roots in early social networking and community communication services, including the bulletin board systems (BBS) of the 1980s, then the Usenet newsgroups, and Geocities in the '90s, whose communities organized around topical interests and provided their users with either email or chat room communications. The worldwide information communication network known as the Internet gave rise to a higher-level networking: a global web of connections among like-minded individuals and groups. Although the basic idea of connecting people across the globe has changed little since then, the scope and influence of social media services have attained never-before seen proportions. Although it's natural that a large part of the conversation is still happening in the "real world," the shift toward electronic information exchange on the level of human interactions has been getting stronger. The proliferation of mobile devices and connectivity puts the "Internet in our pockets," and with it the possibility to get in touch with our friends, families, and preferred businesses, anytime, anywhere.

No wonder that a myriad of services has popped up and started serving our needs for communication and sharing, which led to a transformation of public and private life. Through these services, we can immediately know what others think about politics, brands, products, and each other. By sharing their ideas privately or anonymously, people have the choice to speak their minds more freely than they would in traditional media. Everybody can be heard if they choose, so it's also become the responsibility of these services to find the needle in the haystack of people's contributions, so to speak, in delivering relevant and interesting content to us.

What's common to all these services? They are dependent on us, as they're only the mediators between humans. This means that in a way the mathematical regularities that we may discover through analyzing their usage data reflect our own behavior, so we can expect to see similar insights and challenges when we work with these datasets. The purpose of this book is to highlight these regularities and the technical approaches that lead to an understanding of how users of these services attend to them through the lens of the data these services collect.

Human Interactions Measured

Social media, as its name suggests, is driven by social interactions around the content that the online service provides. Social networking, for instance, makes it easy for individuals to connect with each other and share pictures and multimedia, news articles, Web content, and various other bits of information. In the most common usage scenario of these services, people go to Facebook to get updates about their friends, relatives, and acquaintances, and to share something about their lives with them. For example, on Twitter, because the follow relationship doesn't have to be reciprocated, users can learn about what any other user thinks, shares, or communicates with others. With LinkedIn, a professional social network, the goal is to connect like-minded professionals to each other through its network and its groups, and to serve as an interface between job seekers and companies looking to hire.

There are other social media services where the networking aspect of social interactions is used more as a facilitator rather than an end to co-create or enjoy shared content (for instance on Wikipedia, YouTube, or Instagram). Although the connections among users can be present, their purpose there is to make content discovery manageable for the users and to make the creation of content-for instance, Wikipedia articles-more efficient.

Of course, there are many other social media sites and services, usually targeting a specific interest or domain (art, music, photography, academic institutions, geographical locations, religions, hobbies, and the list could go on), which just shows that online users have the deepest desire to connect to people based on their shared interests or commonalities.

One thing among these services, their vastly different areas of focus notwithstanding, is common: They exist only because their users and audience are there. This is what makes them different from "pre-created" or static Internet locations such as traditional media news sites, company home pages, directories, and just about any Web resource that is created centrally by a relatively small group of authorized content creators ("small" at least in comparison to the crowds of people that use social media services with numbers generally in the millions). The result of the collective dynamics of these millions of social media users is what we can observe when we dig deep into the usage patterns of these services, and this is what we're interested in understanding in this book.

Online Behavior Through Data Collection

When we collect usage log data from social media services, we have a glimpse into the statistical behavior of many human beings coming together who have similar motivations or expectations or act toward the same goal. Naturally, the way the given service is organized and how it highlights its content has a great influence on what we'll see in the logs about the users' activities. The access and usage logs are stored in the databases of the service, and, therefore, the statistical patterns by which we all interact with others and the content the service hosts are bound to show up in these traces. (Provided there are such patterns, and we don't just carry out our daily activities in a completely inconsistent and random way! We'll see that-as perhaps expected by common sense-statistical regularities are abundant everywhere.)

Fortunately, the services (in most cases) don't differ so radically from each other in their designs that they would give rise to completely different user behavior characteristics. What do we mean by this? Let's say, for example, that we want to measure a simple thing: how frequently users come back to our service within a week and take part in some activity. This would be just a number, ranging from 0 to (in theory) infinity, for every user. Of course, we won't see anyone undertake an infinite number of actions on our service within a limited amount of time, but it may still be a large number. So, having set our mind to measuring the number of activities, can we expect to have different statistical results for two different systems: users posting videos to their YouTube channels and users uploading photos to their Flickr accounts?

The answer, obviously, is a resounding yes. If we looked at the distributions of the number of times people used either YouTube or Flickr, respectively, we would, of course, see that the fraction of YouTube users who upload one video per week is different from the fraction of Flickr users who upload one image per week. This is natural, as these two different services attract different demographics with different usage scenarios, so the exact distributions, consequently, will be different. However, what is not perhaps straightforward is that in most online systems that researchers have looked at we find a similar qualitative statistical behavior for these distributions.

By "qualitative" we mean that although the exact parameters of the usage model may be different for the two respective services, the model itself, through which we can best describe user behavior in both systems, is still the same or very similar between the services (with perhaps slight variations).

The good news about this is that we can be reasonably confident that what we're measuring with the data in the activity logs is indeed the underlying human behavior that drives the content creation, diffusion, sharing, and more, on these sites. The other piece of good news is that from this we can extrapolate and if we encounter a new service operating on user-generated content, we can make educated guesses about what we can measure in it. Therefore, if we see something unexpected in the graphs that's different from the general pattern we have seen before, we should look for a service-specific reason for it that we can be inclined to explore further.

So, in a way, the methods and the results that we highlight in this book may well apply to a completely new service if it's also governed by the same underlying human behavior. With few exceptions, this is true of the social media services for which we're aware research exists, and therefore we like to think of these systems as providing insight into human behavior. The opportunity, then, to observe and describe many people acting loosely together is unprecedented; this is because of the digital footprints they leave behind in the services' logs. (Privacy issues are, of course, a valid practical concern, but here we're interested only in the large picture and not how specific individuals behave.) The next sections look at what kinds of data can be of interest in various social media services and which public datasets we'll be using for examples in this book.

What Types of Data Are Essential to Collect?

The questions we would like to ultimately answer with data determine the types of data you need to collect, but in general,...