Press Freedom, Bulent Kenes & Mohammed Rasool — coverage on Twitter during G20 Turkey.
During the G20 Conference in Turkey, which was held on 15th/16th November, we collected all Tweets (in English) related to the conference.
The objective of this post is to illustrate how analysis of community structure within Twitter conversations allows the discovery of interesting stories which are obscured by the issues which tend to dominate the news.
Overall we collected around 300k Tweets over the three days of the conference.
In this previous post there is an overview of the methodology for collecting and analysing Twitter conversations and community structures formed by the groups of people who communicate with each other.
Using the techniques described in that post the Twitter traffic was analysed to detect the communities that formed during the conference. The map below shows one view of the output from this process. Each dot represents a Twitter User and each line a communication (Retweet, Mention) between Users. The colour indicates communities of Users and the size of each node represents the Users importance in the network.
The map is quite interesting to navigate and pick out the discernible features such as important Users and interesting communities. The map serves to illustrate a very high level view of the structure of the conversation. In reality, however, there are many thousands of communities analysed over many iterations which is impossible to visualise in a diagram and any visualisation is only a high level approximate representation.
In order to find usable information from this analysis it is necessary to drill down into the topics of conversation within the communities that are detected.
To put this into context we can start by looking at the overall volumes of Tweets and topics of conversation as signalled by the use of hashtags.
The excellent G20Live shows a realtime dashboard of the volumes of Tweets and important hashtags along with the most important and visible Users. If you want to know the big headlines then this is the first place to look.
G20Live is the kind of resource that provides us with a background level of information which we use to seed further analysis using the community detection process illustrated in the map and a iterative analysis technique known as OODA Loop.
The information from G20 plus is an example of the implicit guidance at stage 1 shown in the diagram below, we use this to direct our initial collection and analysis.
The post referred to above provides further information about this process.
A more detailed analysis
Recall that the objective is to find interesting stories and Users within the Twitter conversation, specifically:
- Find stories and topics that are less obvious than the major headlines.
- Find stories and topics which were not possible to find by conventional search tools where it is necessary to define search terms.
- Find the Users who are influential within the topics and areas of interest.
Twitter Tribes Tool
We have developed an analysis dashboard tool called Twitter Tribes for analysing the results of the community detection and topic discovery process.
Tribes are groups of users who share a common topic of interest. Tribes are detected automatically by machine learning analysis of the communications between users which detects communities as in the map diagram above. Over time these communities coalesce into more permanent groups which are called Tribes. Tribes are named after the most important User when the Tribe was first created, these Users are called the “Mayor” of the Tribe.
The remainder of this post will focus on using a lightweight version of this application which you can view on Tableau Public here — Twitter Tribe Tool.
The initial dashboard looks like this:
There is a lot of information contained in this dashboard, we will concentrate on a simplest use case (and the type of analysis which would be conducted first in practice) for finding non obvious information.
The G20 is an opportunity for activists and interest groups to get coverage for issues of global importance, but which are not always easy to keep in the public eye.
Unfortunately, during this G20 that task was made more difficult by the events in Paris which resulted in a show of unity amongst participants and a high proportion of Tweets on that subject.
However, using the Twitter Tribe techniques it is possible to find other interesting subjects and the people with influence in those areas.
One of the first ways to evaluate this dashboard is to scan the Topics lists.
The first thing to notice is that the topics broadly make sense in the context of Tweets about the G20 conference.
Secondly, notice that the obvious topics indicating the high level main themes are not present. The analysis filters out terms that are present in a high percentage of Tweets. Words like “G20”, “Turkey”, “Obama”, “Putin” and unfortunately in this case “Paris”, “Terrorism”, “ISIS”.
Removing the obvious headline topics makes it possible to discover some more interesting stories.
Amongst the topics that you would expect to see (climate, refugees, finance etc) some are less expected and worthy of investigation. Highlighted in red is the work “rasool” which is clearly not a common word or one expected in this context.
To discover some more about this topic click on the Topic in the list. Now the Tribes list on the right is filtered to just the Tribes that were tweeting about this topic. The Hashtags and Users are also filtered.
Highlighted in green are some elements which start to provide an understanding of this topic.
- Tweets about human rights campaigners calling for release of “Mohammed Rasool”
- Amnesty and Vice News are prominent Users in the Tribes and Users lists.
To confirm what the topic is about hover over the grey topics bar in the Tribes List. In the context menu that appears click the link “Topic & Tribe Search”. This opens a new window with a Google search containing the Tribe name “AmnestyUK” and the Terms in the Topic list.
In this case it is easy to find information about the Journalists who were set free from Turkish prison but one of their colleagues remains incarcerated.
Likewise hovering on the Tweets list shows a context menu to click directly to the Tweet on Twitter.
Having discovered that the topic is about the story concerning journalistic freedom and a specific case of alleged abuse there are other topics in the list which appear to be related.
Highlighting them further refines the lists of Tribes, Hashtags & Users as shown above. Now there are seven Tribes who are identified as discussing these topics.
Using the same process outlined above for searching Topics and Twitter Users it is very easy to identify another specific case of press freedom issues in Turkey.
Finding Influential, Interested and Interesting Users
Returning to the Dashboard and quickly filtering for each of these Tribes and evaluating the Tweets within the Tribe shows that the last one “TheCanarySays” is not really as relevant as the others in relation to this Topic.
Select the six interesting Tribes and then select the User Detailed List tab shows a full list of the Users in these 6 tribes.
The list of Users can be filtered and sorted in various ways e.g. Followers, Reach, Retweets etc.
A simple initial look at the Users identified in this way shows that some of the influential and interesting Users on this Topic have been identified.
Conclusion
This analysis demonstrates that the use of automated community detection algorithms within the OODA analytical framework can quickly identify the important Users and Topics within a much broader Twitter conversation.
The entire process outlined above was conducted within 10 minutes by someone with no prior specific knowledge of the topic.
If you would like to know more about how these techniques could be applied to other areas of research please follow me on Twitter and get in touch.
