Worked on a number of engagements at Capgemini, one of which I am going to talk about was for a Fast Consumer Marketing Goods Company, they wanted to improve their social media analytics suite, and in particular focus on influencer marketing
Identifying the most influential person can often be a difficult task: there is so much criteria to choose from. Do I love the person, do they make me laugh, are we compatible? Who is the one? Contrasting this to digital marketing campaigns – we often need not just one, but many influencers, these influencers are however not for life, but there is different criteria to choose from – do the influencers reach/ engage the number of people I’m looking for? Do they reach the target audience?
Create a piece of software in this case a plugin in Dataiku which can be distributed
Initial data sources are boxed, but additional data sources are being sourced to enhance the service
Wide variety of metrics to get a true 360 picture of an influencer
Lets have a look at these indices in more depth, what they are and why they are valuable
H-index originally came from academia – whereby you measure the influence of an academics publications
On twitter I have defined the H-index to be:
Used in academic literature to measure the ability of an academic to publish papers and get them cited. In the users last 200 tweets, it is the number of tweets that have been retweeted X times 4th tweet, has only 3 retweets -> H-index = 3 Likewise, the H-index can be applied to facebook where you take the last 200 posts and calculate the number of likes on each post
1-request to api per user
[2341, 540, 249, 142, 152, 217, 222, 227, 66, 80] whilst stephen has a large number of retweets, he has only tweeted 10 times! Illustrating he has very influential tweets, but is not very active on twitter!
Peter has retweeted thousands of times, but has been retweeted not very often!
M-index reflects how bursty an influencers tweets are and how influential they tweets are.
Are they tweeting a lot over a short period and receiving much engagement, do they tweet a lot but receive little engagement, or does the candidate tweets very little and have little engagement
G-index reflects not just one tweet but a group of tweets and looks at the distribution of a tweeters retweets
Ideal scenario, is someone like Jamie Oliver, but we can’t always afford such expensive chefs
Fine, maybe we have these two influencers in mind, but how well does the H-index/engagement stack up relative to other Foodies on twitter? Chunk of people who could potentially be more influential & cost less that we can mine into
Bin Width = 20; Bimodal for women, thresholds -> create benchmarks from previous campaigns etc
Bin width = 2k, Most of Denise’s last 200 tweets were done over a 1-day period
Bindwidth = 5
Exponential Curve – Shows that people care more about eating, less about beauty (less engagement), and it drops substantially!
Binwidth = 5; females have more engagement than males, but for beauty there are less tweets that get retweeted a lot
Pydata influencer validation
5th September 2017
Dr Ed Cannon