This is more of an idea, at this point, finding the “Sophisticated” or “Intelligent” customer using Alterian/Techrigy/SM2; the SM2 platform, unfortunately, doesn’t make it easy to export data neccessary to do this simply – but it’s doable.  Is it accurate?  I don’t know – it’s never been tried before, to my knowledge.

Here’s the situation – your a business that caters to a level of sophistication – but how do you isolate that?   Proxies for “sophistication” might be “education level”, “income level” etc (you can get that from Quantcast if you really want it).   Another measure of sophistication might be “emotions” – specifically, emotions surrounding a specific profile that is focused on an business or of an event nature (for PR purposes).

Most Social Listening platforms will break down sentiment to “positive” or “negative” with about a 60%-70% accuracy level (last I heard) and a few will do “Tone” (ie: strongly positive, strongly negative, etc) and one of them, I know of,  attempts to do “emotions” -  Alterian/Techrigy/SM2 – the problem is – we don’t know what emotions correspond to the customers we might want to reach.

In this case, I examined people who might be interested in a certain type of restaurant experience  – and if I looked at some of the results and identified “sophistication”, for example, maybe I could take all the results and sort them for customers who expressed a certain combination of emotions, and then “segment” those results and call it “sophistication”.

This would be trial and error, though the main problem – Alterian doesn’t export results on “Content Emotions“  (there is a work around in Excel using Web Queries, but its ugly and very time consuming to pull the data this way) nor does their comprehensive reporting include the emotion breakdown – and it should.

If Alterian/Techrigy/SM2 could export the Emotions report in this format – you could select “buzz” you liked and see if there was a pattern by just browsing results – then I could sort my “emotions” table, above, and get combinations.

What I want to see – in short – say I defined, by looking a series of results that listening package picked up that my ideal “customer” has the following profile:

I sort my results to get anyone who fits in with these settings – keep on testing and refining the “master table” or “lookup key” till I’ve defined setting that consistently give me decent results.

I’m not claiming this kind of “emotion” report is accurate – though I suspect a lot of the quality of this kind of reporting will depend on a few factors:

1. “Tightness” of the profile – how specific you’ve set up a profile so you get only relevant results in the first place.
2. “Nature of the query” – I found food and consumption easier to gauge in terms of emotion – than, say, complex situations, like an election, or health care, or energy, etc.  The more “grounded” the query, the more likely that emotions will equate to feeling we have, say, in our “gut”.
3. “Accuracy of the emotion algorithm” – it’s been noted by me, and others, that positive/negative and emotion detection in content is pretty tricky to gauge  well – my guess is that people don’t normally say what they mean, anyway, and often emotions are conflicted – so we’d assume that what we’ll see is a number of emotions that are present – we just want to get the right ratios.
4. Traffic of the source url – right now, Alterian uses Alexa while Radian6 uses Compete, which is probably a better source – but neither can deal with individual urls – say a blog post – or tweet – they can only deal with website traffic, and that’s an estimation – usually not the real traffic numbers.