Send the link below via email or IMCopy
Present to your audienceStart remote presentation
- Invited audience members will follow you as you navigate and present
- People invited to a presentation do not need a Prezi account
- This link expires 10 minutes after you close the presentation
- A maximum of 30 users can follow your presentation
- Learn more about this feature in our knowledge base article
Do you really want to delete this prezi?
Neither you, nor the coeditors you shared it with will be able to recover it again.
Make your likes visible on Facebook?
Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.
Twitter Influencer Analysis
Transcript of Twitter Influencer Analysis
Along with rankings it shows Topics...
Let's select just UN_Women...
Major Change to Conversation...
Let's look at what happens to our Observe process if we do nothing.
View of the Battlefield
Recall our OODA process originates in military strategy. So let's use the battlefield analagy when we look at the network map..
Implicit Guidance for
Data Science & Big Data...
Start with a wide search for terms related to the topic of Interest
(datascience OR “data science”) OR (bigdata OR “big data”) OR ( data AND (algorithm OR viz OR vis OR python OR SAS OR SPSS) ) OR (machinelearning OR “machine learning”) OR rstats OR analytics OR “data mining” OR “artificial intelligence” OR AI
Twitter SNA Analysis
Find Influential Users on Twitter
Social Networks are Messy and full of Noise...
Use Community Detection to discover the self organising groups in social media
: Data Science & Big Data ...
Over a continuous period of 10 days (14 Sept — 25 Sept) the search collected 600k Tweets...
The interesting traffic gets drowned out by the spam originating from TweetBots
However there is a problem
As mentioned earlier social networks are messy and full of noise...
So the first Decision:
Remove Tweet Bot Noise..
Some of these Bots are easy to identify by human inspection...
Others are more sophisticated
They do NOT contribute any value as they just repeat other people's content
First Step in the Process is the initial iteration of Observation guided by our existing understanding of the domain
Let's Look at the Process...
We ran our first Observe phase.
Analysis revealed a lot of noise..
Therefore we need to adjust the Observe process
Remove all Users with small number of Followers
In this case we only keep Users with more than 200 Followers
Other use cases e.g. consumer brands we may want to keep these smaller Users
To Remove Tweet Bot Noise..
However, after few iterations of this kind
we can produce a list of the most
influential Users in the world of
Data Science & Big Data
These are the individual Users but what
about the communities they influence..
To illustrate further let's look at a different example from the domain of Climate Change.
Using the cycle of Observe and Orientation over a period of weeks we collected the Twitter Conversation about:
Climate Change; NGO & Not for Profit; and Philanthropy
Community Detection algorithms create the communities which are highlighted in the different colours.
Here is a list of some of the members of the UN_Women Tribe...
Just from the descriptions and names we can see that there is a very strong core of Users interested in Gender Equality Issues.
Here we can also see some of the topics of conversation within the Data Science Communities
After we establish the tempo of the cycle we can start to see the emergance of significant communities...
Lets look at the community shown in yellow at the top left..
Notice one significant User - UN_Women..
Sometimes the network map vizualisation is useful in identifying the members of the community.
However, sometimes the communites are small and hard for a human observer to spot..
Each time a community is detected it is named after the most important User in the community measured by Page Rank. In this case @UN_Women
Each day the algorithms are run several times identifying thousands of seperate communities.
The members that are in the same communities are collected together into a Tribe.
Over time Users with the greatest affinity with each other appear in the same communities many times.
Here is a list of the Tribes that were identified as sharing a conversation about Climate, NGO & Philanthropy issues.
UN_Women ranks pretty highly.
The Tribes are ranked by Page Rank(Orange)
The other measures are:
More on these later
Let's look at what defines a Tribe...
Here is the dashboard from which the chart was taken
We can also see that some very specific hashtags are featured prominantly in the list.
This period was the first day of the COP21 conference - we can see that the main topic detected was about the conference.
If you are wondering about Iceland...
So, we can see that Tribes have an identifiable shared interest.
Let's look at the people within the Tribe.
We need a
Remove all Users on a Black List of Users
The Black List is built up over time by manual identification and/or automated machine learning
Re Tweet most of their content
Get Low numbers of Re Tweets and Mentions from other Users
Tweet a lot & in bursts
Not Verified by Twitter...
In practice more steps are required over several days or weeks.
More subtle spammers work in networks like these
Some genuine users use similar tactics to spammers
But what happens when there is a
major change in world which affects
the conversation on Twitter
DAY BEFORE START OF COP21
We can already see in the Topics that there is a story about the 'Thousands Marching' that dominates the conversation on this day
DAY ONE OF COP21
The Conversation is now totally dominated by a single Topic of Conversation
Now we have a situation where a major event is skewing the analysis.
There are two ways we would adjust to this phenomenon...
Adjust to events within the process...
COP21 is something we know about in advance so it is an input into our process as Outside Information
Alternatively our Analysis process detects the change as New Information
In either case the adjustment needed in this case is a list of words which are associated with the major event.
Here is a snippet of the Stop Words used in this case.
Words which are used in almost every conversation are adding no information. They are purely indicative of the overall Topic of interest - in this case the COP21 Conference.
When these words are removed it is possible to see the Topics at a more granular level.
The decision about which terms to filter is a judgment call dependent on the use case for the analysis.
In the case of COP21 it is judged that the 'thousands people march' Topic is a material part of the story that day.
So, finally here is an example of how the visualisation can help with making those judgements.
So, we adapted one from the world of military strategy
The 'battlefield' view of the COP21 conversation looked like this...
Let's contrast that with the view of the conversation from a different domain - The Ebola Crisis...
Ebola Crisis 2014...
This has a similar structure to the COP21 map...
However, that is only part of the story...
The distinct communites generated a significant volume of content...
A visualisation like this can help in gaining insight into the overall structure of the conversation
So, running this process takes us from this...
Our initial knowledge
Making Sense of it
By identifying communities we can discover common interests amongst Users...
Words which appear only in certain communities can be easily identified as relevent or not...
TCOT, Rednation, GOP, Obamacare, Benghazi
Gun Control, Prepper, Securetheborder, ban flights
We are looking for people like this
We are looking for experts and opinion formers with real value and expertise to contribute to the conversation..
Here we can see some of the major players in the COP21 conversation - the colours indicate the way they group together to form comunities..
World famous people or companies are easy to find. They control the major news stories - the challenge is to find those interesting Users with value to add but not the power to dominate whole conversations..
In a big network there are a lot of potentialy interesting people
The first time we feed forward the observations into the Orientation Phase and examine them they look more like this...
The grouping and colour of these maps helps us see some of the way Users group into communities..
Once we have identified the communities we can start to discover other information...
To find them...
Key lesson from military analogy - Orientation...
creating, employing and dealing with the novelty that permeates human life (Boyd, 1992)
Using these techniques we can identify who the important influencers are and who they influence in your domain of interest.
There are many use cases for this information:
Engaging with key influencers
Research and new information - finding experts
Using influencers to amplify your message