Social Media Network Analysis and… Grapes?
Grapes (image from http://en.wikipedia.org/wiki/File:Thompson_seedless_grapes.JPG) This is clearly a social network. Obviously. |
My Guide to Eating Grapes and Analysing a Brand's Social Media Community
Ok, so I may be losing it a little. Or a lot. I've been doing network analysis today, using Gephi and the Force Atlas 2 layout algorithm. Using data collected from Twitter, I was looking at all users who tweeted and mentioned a particular charity over an 8-month period to gain insights into how the network of supporters was formed. What I didn't expect to be doing was searching for images of grapes.
First, the details. I imported the data into Gephi (the University Tweet Harvester handily outputs Tweet data in graphml format, so this was trivial) and began by removing the charity in question's own node (and any related edges) from the dataset. This step is important as it removes the 'ego' from the network. When visualising the interactions around and about a brand, topic or organisation, including the organisation's account in the image can cause very distorted images as everything will be linked to that node and will hide other important features.
I then calculated the in-degree and out-degree of each remaining node. In-degree represents how many times that node was mentioned in a tweet containing the charity's username. Out-degree represents how many tweets that user made mentioning the charity's username. Nodes were then modified to reflect these values, Gephi automatically ranks nodes and so higher out-degree nodes were made larger, and higher in-degree nodes were made 'darker'.
I then ran the Force Atlas 2 layout algorithm on the network so that nodes connected to each other (i.e. people who had mentioned each other) were moved closer to each other, and further away from those they weren't connected to.
The result was an image (shown below with labels removed) that resembled a large, central cluster, a scattering of lone nodes that were connected to nothing else, and some small clusters around the edge.
![]() |
A Social Network Visualisation of people discussing a charity on Twitter over 8 months. |
Basically, a community network is a collection of 'bunches' or clusters, often containing one dominant and many smaller collections. Much like in a tub of grapes. There are also a large amount of lose nodes - those grapes that have fallen from the bunch and roll around in the bottom of the box. So, what does this mean?
The largest 'bunch' often contains the larger nodes (this isn't surprising as it is likely that if you're talking to other people about a topic, it is more likely you'll end up sending more messages about it than if you're just posting a comment about something to nobody in particular). The largest grapes in a bunch are always the most attractive and desirable, and would be the ones to start with for eating. This is similar in a network (except I am not claiming here that the most vocal people in a social network should be eaten, this analogy has to stop somewhere). The most communicative people in the network, and the most dominant cluster is most likely a significantly important collection of users discussing the brand, and would therefore be worth looking at first. Remember, these sizes are based on tweets mentioning the charity, but not directed to the charity, as that node was removed. So these people are discussing it, independently of the organisation itself. Brand ambassadors. Enthusiasts. Engaged supporters.
Next, the small bunches. If you're offering some grapes to someone, they may take one of the smaller bunches, not wanting to just pick a single grape (they could just pick a loose one for that, but these always seem to be overlooked and those that are still connected to something else are favoured), but not wanting to take the majority of them by selecting the biggest cluster. So they take the small bunches around the side. Again, with a community network, these small clusters are the interesting, attractive groups of users that are slightly separate from the main community. They're easy to access and see what the conversation was about because they're smaller. You could say they're the low-hanging fruit for more qualitative analysis of what is being discussed.
And then there are the lone, isolated nodes. In a tub of grapes, there always seems to be a load of these at the bottom, those that have fallen and gradually been forgotten as you focus on the larger bunches. In the diagram, they represent the users who have mentioned the charity a very few times (generally just once) but have not been mentioned themselves in any other tweet about the charity. They could have been replied to by the charity, but this is not shown in the network for the reason discussed above. They're not to be forgotten - everyone on social media has a voice - but to analyse them individually is more time-consuming than simply lifting the entire bunch out of the tub. Also, no matter what quality grapes you get, there will always be some that are a little squashed or wrinkly. The undesirables. They might represent the angry people on social media, the grumpy ones who cause problems and need to be watched out for.
Right, to summarise:
Social network visualisations of brand communities are like boxes of grapes. The largest, juiciest bunch is the nicest, and should be picked first for eating (analysis). Assuming time and resources allow it, the smaller bunches should then be pondered over, before moving onto the stragglers - ensuring that any 'bad grapes' are dealt with accordingly (by this I mean responded to in a positive way or some other form of non-confrontational response, not with rudeness or aggression!).
More...
Where this could go wrong:
The largest bunch could be a negative cluster who are complaining about your organisation. In this case, you have a lot of bad grapes, and need to figure out why. I'm sure there'll be some good ones in the box somewhere, you just need to find them, maybe a small bunch of them, and then help to amplify their positive message.