I’ve been playing around trying to get Page Insights data directly from the Graph API (usually we just export and process the data from the Insights dashboard.) Thanks to an excellent article by Facebook’s Paul Carduner on Authenticating with Facebook on the command line, I’ve found a way to do it; but if you just need a short term token (for an hour or two’s experimentation), then it’s actually quite simple to get one from the Graph API Explorer tool. Here’s my quick guide.
I feel I tend rather to go on about this — but it bears repeating. Your Facebook audience isn’t connecting during work hours on weekdays and there’s a strong chance that they’re not on a desktop PC.
I gave this version of the talk today at the Social Media Marketing 2012 conference. It’s been an excellent and instructive day — many thanks to Luke Brynley-Jones & Our Social Times for putting it on.
I’m sorry that the fonts we use don’t really come out well on SlideShare; but you should get the gist. The deck is downloadable — please feel free to use any bits you want, and I’ll happily answer questions over at @mediaczar
Links to some detail on the case studies mentioned
I’ve been in training with Altimeter Group’s Charlene Li and Ed Terpening for the last couple of days. One of the case studies that came up in passing was the US department store chain Kohl’s Facebook Page (more than 9 million likes at time of writing.)
Here’s a quick flow map of the last 2.5k posts on the Wall between 18 August and 23 October this year.
We can see that there are ~2k peer responses to UGC posts (the yellow circle) or around 0.8 peer responses per post. This is a very high number: not what we’d expect to see. Continue reading
Another chapter in the “like if you hate cancer” story. Interestingly, user posts on Lil Wayne’s Page don’t seem to be visible to other fans; so did the initial impetus for this post come from Lilquan’s own social graph?
There’s a Facebook property,
page_fan that I’ve been looking at in the hope that I can begin to unpick this kind of question: “does peer-to-peer activity on UGC Page Posts come mostly from other followers, or from friends?”
I did a little more digging. As I might have suspected (given what I’d already seen of this meme on Coca Cola’s Page) Lilquan wasn’t the only user to spam Lil Wayne’s Facebook Page with a “like if you hate cancer” post, just one of the most successful to date.
This chart shows the likes on ~1.5K user posts containing the keyword “cancer” on Lil Wayne’s Page between 27 September and 12 October. Note that four of them (including Lilquan’s) have broken the 100K likes mark.
I’ve posted the data on Google Docs as an Excel spreadsheet
There’s no rubric, but I’m assuming that this is a form of network graph — where each node may itself become the source of new edges.
What’s interesting, if that’s the case, is how only a few shares lead to huge bursty cascades of sharing. I’ve seen this before on much smaller data sets:
There’s a potentially useful paper, The Dynamics of Viral Marketing, (Leskovec et al., 2007) that addresses this phenomenon. I’d be grateful for links to other papers — or your thoughts.
Thanks to somerandomnerd for the link.
I’ve collected over 1,100 posts from Coca Cola’s Facebook Page that follow (more or less) this format:
Coca cola, I have a proposal. You have 9million followers if I can get 2million likes, you have to change one of the colors on your cans to pink. And you have to donate 30% of the profits to research for the cure of breast cancer.
So people like it up!!! (source)
Nearly all of them have been posted in the past week. The chart above shows (on a log scale) how many Post Likes each of them has received.
Min. 1st Qu. Median Mean 3rd Qu. Max.
0 2 10 1655 49 360000
There’s a little bit of evidence (the “9million followers” error — Coca Cola had more than 52m followers when this was posted) that this meme has jumped here from another Page.
Coca Cola and Cancer
There are a few of these going around at the moment (click on the images to see the original post)
Austin has collected over 300K slacktivist likes already: even though he can’t count too well — how many fans does Coca Cola have?
I’ve spent my free time this weekend fiddling around with perl and Gephi to see whether I can eke any additional meaning or insight out of the Facebook data that we collect.
The chart shown above (you can download it from here) is a visualisation of all interactions between members of the public on Marks & Spencer’s Facebook Page. Each node represents a member of the public, and the lines between them represent comments on each others’ posts. I’m being a bit careful about my language here: they’re not necessarily followers (or fans) of the M&S Page, it’s just as likely in most cases that they are Facebook friends of the original poster (or O.P.)
To identify potential areas of interest, I’ve mapped both size and colour to the number of incoming commentators (indegree is the useful technical term here.) So an O.P. who’s represented by a large orange or red dot has attracted comments from lots of other people. The thickness of the lines is proportional to the number of comments made by an individual commentator — so if there’s been a bit of back-and-forth, the line will be thicker.
I’ve identified four stories of interest in this way; two negative, and two positive.
7 May 2012: (big red blob at six o’clock) Granddaughter complains that her grandmother has been unfairly victimised for forgetting to pay for a rose bush. Boycott threats. The works.
22 August 2012: User complaint about (lack of) school uniforms heats up.
28 August 2012: (pale orange blob at half past two) Mother petitions M&S to feature her learning disabilities son in a Back-to-School or Christmas campaign. The company quickly meets the challenge and he’s included in the M&S magazine.
10 September 2012: (orange blob at twelve o’clock) M&S helps child replace missing teddy bear. Heartwarming stuff.
I’d say that the M&S community managers are earning their keep.
I’m surprised at the moment how much interplay there is between members of the public. As I say, I don’t know (and I think it’s hard to tell) how many of them are actually fans of the Page, and how many (as in the uniforms debate above) are simply friends of the O.P.
- The R Project (software and manuals)
- John M Quick’s R Tutorial Series
Computing for Data Analysis at Coursera.
- the R-Podcast
- Cookbook for R
- Quick R
Perl and R
- Machine Learning for Hackers (O’Reilly) (££)
- L Francis: Text Mining Handbook – Casualty Actuarial Society
ggplot2 (for pretty charts)
- Hadley Wickham’s ggplot2
- Tony Hirst’s Ouseful.info
I’m learning at long last how changing scales can expose patterns. Today I’ve been looking at the Facebook Page of Marks & Spencer (a UK retailer.)
Using a linear scale on the y-axis, we see how engagement has increased rapidly over recent months. That’s pretty much text book stuff for a well-run Page.
But by dropping a log2(x) scale onto the y-axis (why log 2? I was experimenting after reading When Should I Use Logarithmic Scales in My Charts and Graphs?) we can see a couple of strange patterns emerging.
Notice the distinct vertical lines around October and December 2011, and again in February 2012? Also the horizontal lines that extend from around January to April 2012?
Both can — it turns out — be explained by odd posting behaviour related to photo albums; and could raise some interesting issues of what is and what isn’t best practice.
But the short point is; I wouldn’t have seen it but for the log scale.