Naughty or Nice

Idiro Technologies can reveal that its powerful SNA Plus technology has been used to help Santa with the massive task of sorting well-behaved children from naughty children, prior to the annual distribution of presents to good children at Christmas.
For centuries, Santa has faced the daunting task of deciding which children had been nice (and would receive the present of their choice) and which had been naughty (and would receive a bag of coal or nothing at all).  Through the year, Santa’s research elves collect data on good and bad behaviour by children, resulting in a limited good / bad child dataset. This data is supplemented by a large dataset of Santa letters each Christmas.  However, a proportion of these letters have been found to contain significant inaccuracies.
This year, Santa has decided to use analytics to improve his accuracy in determining the naughtiness or niceness of his customers.  Under an exclusive agreement, Idiro’s telco customers worldwide have given Santa permission to use their call detail records (CDRs) for a unique project to benefit the world’s good children.  

Using Idiro’s SNA Plus technology, we have built a customised social model of the world’s children, including all their social links.  Santa provided Idiro with access to his partially-complete database of known good and bad children. It is well known that child behaviour is homophilous – i.e. that good children tend to associate together, while bad children usually run with a bad crowd. Idiro uses these principles, along with Santa’s partial good/bad database, to develop a Social Naughtiness/Niceness Score (SNNS) based on the known behaviour of the child’s peers. 
In addition, Idiro is working with Santa to improve the text analysis of the millions of Santa letters to identify syntax that shows sincerity or gives rise to suspicion. 
Santa and Idiro estimate that the use of this new technology will result in an improvement in Santa’s Good Child Identification Rate (GCIR) from an average 93% to an all-time high of 99%.  This project is funded by Idiro’s corporate social responsibility programme.
Idiro wishes all our stakeholders the compliments of the season and a happy new year.
Note: Idiro Technologies is a world leader in the use of Social Network Analysis in marketing.  Although Idiro is providing this service pro bono, the Idiro team is looking forward to Santa’s visit on Christmas Eve.   

Posted in SNA

What about the people who do not use the internet

We, the internet generation (if you are reading this you are almost certainly online) sometimes forget that many people do not use the internet. This week I was surprised to read this report, which states 27% of the UK population still do not use the internet. Note that this measure is different to (and I think more useful than) measurements of internet connections or PCs. We ignore all these people at our peril. A poll predicting the outcome of the 1936 US presidential election got the result spectacularly wrong by polling telephone owners. This was in an age when telephones were for the rich, or at least not for the poor. FDR romped home and the pollsters were left red-faced.

When we dive deeper into the internet penetration statistics, we see that many of those that use the internet stay away from our favourite communications media. Perhaps about half of the Irish population use Facebook, though finding the proportion that uses it regularly is more difficult. How much do non-users of the internet spend? Less than internet users, almost certainly – but still, they are an increasingly neglected market segment. How big is that segment in your country?

The mobile phone ownership stats from the ITU paint a compelling picture. Most countries in the world have over 100% mobile phone penetration. While there are still people who do not use a mobile phone, they are far fewer than for nearly any other medium. Many more people cannot read or write, for example. By focussing our marketing on internet users, and in particular on the users of online social networks, we miss out on a large and increasingly under-served segment of the population. When analysing networks to find influencers (and the influenced) to target campaigns, it is far better to work with mobile phone data than any other publicly available dataset.

Interesting stats from WOMMA

The USA-based Word-of Mouth Marketing Association (WOMMA) comes out with some useful stuff from time to time, including the infographic shown below. Of course, it’s USA-focussed and it’s lacking the detailed references that would increase its credibility, but nevertheless, it’s interesting.  For more, browse WOMMA’s resources section or show up at their conference next month.

With thanks to Jackie at Church of the Customer.

WOMMA infographic

Alcohol does not cause promiscuity and violence; your friends do

Or so anthropologist Kate Fox claims in this BBC article. The evidence appears persuasive – in some cultures, when people get drunk they behave in ways that they would never do sober. In others, even very drunk people do not lose their inhibitions. Therefore, she argues, we should not blame the alcohol per se, but in fact blame the way in which our peers react to it. It’s a fascinating theory, which ties in with the work of Christakis and Fowler, among others, on the influence that our peers have on our behaviour.

Maybe one way to tackle antisocial drinking might be to use SNA to identify the influencers for alcohol-related behaviour, and educate/incentivise them to change their behaviour. Another might be to export antisocial drinkers to cultures where such behaviour is not approved of. Now that would be an interesting, if controversial project.

Use of language on Twitter can identify gender?

Social media sites such as Twitter, Tubmlr and Facebook can anonymize the identity of a user, while at the same time enabling them to relate a wide variety of information, comment and opinion. The sometimes spurious link between on-line identity and the message portrayed can be used to impersonate real or fictional personalities, occasionally providing a seeming credible account of current news and events.

However, a recent paper published by John D. Burger and colleagues suggests that a persons use of language can betray certain aspects of their (hidden) identity, most notably their gender.

Their study, entitled “Discriminating Gender on Twitter”, looks at how randomly selected twitter updates (tweets), can identify the gender of a user to an accuracy of approximately 70% [3].

Aside from garnering unwarranted credibility, self-anonymization can lead to a breakdown of social norms, as disentangling a persons “good name” from their identity has been shown to cause them to act with a lessened sense of responsibility [4]. Although the use of language can suggest certain demographic features, it may provide a useful insight into user behaviour when combined with more traditional social network analysis methods.

Data-mining Facebook

As we all know, one of the largest social networks currently available on the internet is Facebook, with approximately 600 million active users, and upwards of a billion links (as of June 2011). You could do a lot of interesting data mining if you could get your hands on Facebook’s data…and it seems you can. Apparently, almost 75% of Facebook users enable the default privacy that exposes their private data to web-crawlers. These web-crawlers can (and do) hop from link to link in Facebook, making it possible for them to navigate a large proportion of the entire Facebook listing. This is a dataset consisting of 44TB of subscriber and link information.[1]

Once the data has been gathered, it is possible to examine the community structure within the network of connected Facebook users in more detail, and infer behavioural trends between similar users by applying various linear-time and heuristic community-detection algorithms.[2][3]

One of the questions we have to ask here is: cui bono? Who benefits? These outside web-crawlers are not affiliated to Facebook and have not asked Facebook users for their permission to trawl and gather their data. Couple this with face-recognition software which is becoming quite sophisticated and it becomes apparent that our data is quite likely to fall into the wrong hands.

People underestimate the power of data mining and how much can be learnt about a person from the tiniest fragments of data. Careless users of Facebook are ideal targets for identity thieves as they give way too much information away and, once that information is out there, it is very difficult to pull it back.

The bottom line is: don’t ever underestimate the power of data mining. We all want to be better understood as customers, and data mining can really help in that regard, but we should always remember to value our data and to take steps to protect it in this digitally connected world.

[1] – Catanese S., De Meo P., Ferrara E., Fiumara G. and Provetti A., Crawling Facebook for social network analysis purposes, International Conference on Web Intelligence, Mining and Semantics, 2011.
[2] – Leung I., Hui, P., Li, P. and Crowcroft, J., Towards Real-time Community Detection in Large Networks, Physical Review E, 2009.
[3] – Catanese S., De Meo P., Ferrara E., Fiumara G., Provetti A. Extraction and Analysis of Facebook Friendship Relations, Computational Social Networks: Mining and Visualization, 2011.

Posted in SNA

How many friends are too many?

Social Network Analysis is predicated upon the assumption that people tend to maintain links with their friends over time, by calling, texting, sending IM and otherwise interacting over various forms of Social Media. However, there tends to be an upper limit to the number of meaningful social relationships that can be maintained [1], even through the relative convenience of online networking through Facebook, MySpace, Sina Weibo, Studivz or Twitter to name but a few. A recent article in IEEE Spectrum suggests that on average people are capable of maintaining stable social links within up to approximately 150 people [2]. The benefits of Social Media then, are not in the facilitation of greatly expanded, active social networks (e.g., the average number of friends of users on Facebook is about 120), but in slowing down the “rate of relationship decay by allowing us to keep in touch with friends over long distances”.

[1]- Carl Bialik, “Sorry, You May Have Gone Over Your Limit Of Network Friends”, Wall Street Journal Online, November, 2007

[2]- Robin Dunbar, “How Many “Friends” Can You Really Have?” IEEE Spectrum, June, 2011

We are all influencers

As the discipline of social network analysis (SNA) develops and experiences wider adoption, it has become clear, at least to us here in Idiro, that there are a number of erroneous ideas circulating which undermine the integrity of SNA as it is applied to telecoms.

One idea that needs to be addressed is that alpha users or key influencers, if some vendors are to be believed, are the be all and end all of SNA. These vendors believe that all you need to do is target these key influencers with your product and you will experience massive product adoption. It’s a nice story and it may sell well but in reality, it’s just a fairytale.

Perhaps it all began with the publication of Malcolm Gladwell’s book The Tipping Point. In it he identifies certain individuals as being key to the adoption and diffusion of new products/services/ideas. While Gladwell had a point, it has since been hijacked by certain individuals and companies, and sold on to unwitting customers as being the holy grail of marketing.

The idea that there are certain individuals who everybody else follows, regardless of the topic, is very misleading. We are all influencers and we can all be influenced by anyone, depending on the context. Many SNA vendors peddle the notion that the there is a rigid hierarchy of influence amongst individuals that you can discern in your customer data simply by looking at the volume of communications traffic or links between them.

Furthermore they argue that by targeting the individuals at the top of the hierarchy, you will have an exponential lift in your marketing campaigns. Our experience shows that this notion is not grounded in reality. Eight years of research by Idiro has shown that time and again the influencer in a group regarding, for example, a new value-added service is frequently a different person to the influencer for a new phone or a change of mobile phone network.

Our argument is that everything is relative and context-sensitive. Peter may influence John when it comes to picking a restaurant but John may hold sway when it comes to fashion. We all have opinions – some held strongly some not so strongly. We influence those around us but we are also influenced by them.

So if you see a social graph of a community and there is someone at the centre, don’t automatically assume that he/she is the right person to target with your offering. They may be but you need to dive deeper into the data to find out.

For years, Idiro has been helping companies quantify the levels of influence people exert on each other and in turn get significant and measurable ROI from social network analysis. If you have any questions or want to learn more, just email me directly or check out our case studies for more information.

Posted in SNA

Sharing information corrupts wisdom of crowds

Although the actions of individuals are known to be influenced by trends within their social network [1], their opinions and estimation processes may also be affected. In a recent study published by Jan Lorenz and Heiko Rauhut of ETH Zurich, Switzerland, and also described by, the ability of crowds to accurately estimate a particular attribute declined as their knowledge of others’ choices increased.

While additional predictive performance may be achieved when crowd-sourcing a certain question, the divergence of opinions narrows within the group as more information is made available to group members, potentially decreasing any accrued crowd wisdom.

Within the area of telecoms, this study may point the way towards greater accuracy in predicting handset diffusion or churn within mobile operators networks, through an analysis of information flow within a social network.

[1] – D. Kempe, J. Kleinberg, and E. Tardos, Maximizing the spread of influence through a social network. In Proceedings of the 9th ACM International Conference on Knowledge discovery and Data Mining, 2003.

Posted in SNA

Research shows that face-to-face influence is more important than online

We all knew this, right?

The Church of the Customer just blogged on some interesting research by Colloquy (as reported by Where do people most talk about brands? Face-to-face, actually! Even among 18-25s, it’s the most popular medium.

colluquy findingsIn my humble opinion, the reason is largely provided by Robin Dunbar’s ethnographic work on why people chat. Here’s an image. The full report’s available from Colloquy on (free) registration.

For more on how the power of word-of-mouth can help your business, contact