So far, we have analyzed Twitter and tweets to uncover some interesting insights using techniques and concepts of trend analysis and sentiment analysis. We've utilized different attributes of tweets, like creation time, location and even the text itself, to answer certain questions. In this section, we will touch upon Twitter's network aspects. #BraceYourSelves

A social network is a network or a graph at its core. In formal words, a social network is generally a graph representing its users as nodes (or vertices) linked to each other based on certain relationships called edges. Each social network has its own definition of these relationships. For this section, we will focus on Twitter's relationships and network in general.

In Twitter-verse as we all know, there are no friends! A friend relationship is usually a bidirectional relationship, that is, if A is a friend of B, then it is safe to say that B is also a friend of A (well, usually; see Facebook friends). Twitter has tweaked this concept of friendship and allows its users to form unidirectional relationships or follower relationships. On this platform, a user may choose to follow any number of users, based upon their interests. The person being followed though has no such obligation to return the favor, that is, if A is following B, then it is not necessary for B to follow A.

At first this sounds a bit awkward but it is rather an interesting and a very human concept. In our social setting we come across many people about whom we are interested, say celebrities or leaders. In general, we are curious and intrigued by such personalities and love to know more about them. The follower relationship model allows Twitter users to follow such personalities (aka handles) without the personality following the user back. Nothing stops people being followed to follow back, but imagine a Twitter where @KatyPerry would have to follow back all of her 94 million followers!

Very much like other things associated with Twitter (like sentiments, trends, velocity and so on), these follower relationships have also intrigued researchers. As part of this section, let us touch upon the concept of follower graphs and see what magic unfolds.

We use a social network like Twitter to share information about a number of topics and we all want our story to reach as many people as possible. With Twitter, it's the follower count that's most coveted. Popular accounts, such as those of Katy Perry, Barak Obama, Mars Rover and so on, have millions of followers. With this use case let us plot and analyze a given user's follower graph.

To begin with this use case, let us perform our ritual initialization steps of loading required packages and connecting to Twitter using the twitteR package. We use the getUser() function to extract a given user's details and then use the function getFollowers() to extract handles of his/her followers. The following snippet does the same:

setup_twitter_oauth(consumer_key = CONSUMER_KEY,
                    consumer_secret = CONSUMER_SECRET)

# Begin with a certain username
coreUserName <- "jack"
twitterUser <- getUser(coreUserName)

# Extract Followers for the core user
twitterUser_follower_IDs <- twitterUser$getFollowers(
                            retryOnRateLimit=10)
# Typecast as a dataframe
twitterUser_followers_df = rbindlist(
                  lapply(                                                     twitterUser_follower_IDs,
                    as.data.frame
                              ))

Once we have this data, we can proceed towards building a graph. But what fun would it be to have just you at the center of this graph and your followers listed out visually. We make this interesting by going one level deeper, that is, for each of our followers, we extract their list of followers as well.

For building a graph or network, R has another powerful package called igraph. The igraph package provides multiple ways of generating a graph. We prepare a data frame with two columns, that is, from and to. Each row of the data frame represent an edge from one user to another.

In the following snippet, we first filter our followers DataFrame(filtered_df) by removing dead and dummy accounts (you may choose to follow some other criteria or not do any clean-up at all). Then we iterate over each of our given user's followers, extract their followers and append that list to our DataFrame representing the edges:

# extract @ twitter handles of followers of given user
filtered_follower_IDs <- filtered_df$screenName

# prepare initial edge data frame
edge_df<-data.frame(from=filtered_follower_IDs,
                    to=rep('rghv_bali', 
                           length(filtered_follower_IDs)), 
                    stringsAsFactors=FALSE)


# Iterate and extract list of followers of followers

counter = 1
for(follower in filtered_follower_IDs){
  # fetch follower list for current user
  followerScreenNameList <- get_follower_list(follower)
  
  print(paste("Processing completed for:",
            follower,
            "(",counter,"/",
            length(filtered_follower_IDs),")"
            ))
  # append to edge list
  edge_df <- append_to_df(edge_df,
                       list(from=followerScreenNameList, 
                   to=rep(follower,
                            length(followerScreenNameList))))  
   counter <- counter + 1
}

The edge data frame generated using the preceding snippet has each row representing a follower relationship as follows:

Follower/Edge DataFrame

Now that we have our edge data frame ready, we'll prepare an igraph data frame object based on this. After that, we utilize a couple of igraph utility methods to remove loops and multiple edges (if any) using simplify() and make some aesthetic changes to node sizes for clearly marking out our core user. Check out the following snippet:

# prepare network object
net <- graph.data.frame(edge_df2, directed=T)

# simplify networknet <- simplify(net, remove.multiple = F, remove.loops = T)

# adjust the size of nodes based on in and out degrees
deg <- degree(net, mode="all")
V(net)$size <- deg*0.05 + 1
V(net)[name == "core_user"]$size <- 15

# node coloringpal3 <- brewer.pal(10, "Set3")

# overall follower graph
plot(net, 
     edge.arrow.size=0.01,
     vertex.label = ifelse(V(net)$size >= 15, 
     V(net)$name, NA),
     vertex.color = pal3)

The preceding snippet generates the following follower graph:

Sample follower graph

If we analyze the preceding graph carefully, we can see that there are quite a few users/nodes who have edges (that is, follower relationships) amongst themselves! Also, the size of each of the nodes (barring the core user's) is a representation of its follower count. To be precise, we adjusted node sizes to account for the number of incoming and outgoing connections (the in-degree and out-degree of each node). This helps us identify which of our followers are more popular than others.

Even though Twitter uses follower-based relationships, it doesn't stop users from following back. A follower who follows back sounds much like a friend relationship. It would be interesting to identify who our core_user follows back. For a real use case, it might be interesting to identify, for instance, which world leaders follows back who, and derive political insights of sorts from it.

Our graphing package, igraph's utilities simply make our lives easy. It exposes a set of iterators for both edges and vertices along with a utility function called ends(). This function helps us identify whom does the core_user follow back from his list of followers!

The following snippet achieves the task of identifying friend nodes:

# identify friend vertices
friendVertices <- ends(net, es=E(net)[from('core_user')])[,2]

Now that we have our list of friend vertices, let us make some visual changes to our follower network visualization to clearly identify the friends amongst followers. The following snippet marks edges amongst friend nodes as red and enable vertex labels for the same:

# Generate edge color variable:
ecol <- rep("grey80", ecount(net))
ecol[which (V(net)$name %in% friendVertices)] <- 'red'

# Generate node color variable:
vcol <- rep("gray80", vcount(net))
vcol[which (V(net)$name %in% friendVertices)] <- "gold"

plot(net, 
     vertex.color=vcol, 
     edge.color=ecol, 
     edge.width=ew, 
     edge.arrow.mode=0, 
     vertex.label = ifelse(V(net)$name %in% friendVertices, V(net)$name, NA), vertex.label.color="black",
     vertex.label.font=2,
     edge.curved=0.1
     )

The output plot shows the #friendsAmongstFollowers:

Friends in follower graph

The preceding follower graph analysis helps us understand a social network like Twitter from a network perspective. It also shows how different questions could be answered with such analysis and as always, R and its packages help us quickly get to the answers once we have an approach in our minds.