I’ve been doing a great deal of reading on machine learning, collective intelligence and other applications for statistical analysis.
I’ve become one of those mouth-breathing weirdos that enjoys the statistics and methodology portion of their social science classes…and this frightens me. No funny smell or obnoxious verbatim quotations of Monty Python yet!
Here’s a analysis engine that would converge a number of algorithms into a coherent package that would scare the jibblies of a great many politicos.
In this case you would take news feeds, blogs, newspaper editorials, PR documents from relevant organizations and the like and feed them to a computer.
An initial filter would be applied to exclude irrelevant documents - this would be a supervised algorithm similar to those applied to spam filters. Only, in this case it would filter in all articles relevant to climate change.
So we’ve got a big pile of texts relating to climate change the next step is to cluster them in terms of their stance on specific issues. We would then create a clustering machine that groups articles together based on rhetorical usage. So articles written using the same keywords, phrases, sentence structures. It would, by its nature also group those documents which are similar in stance on particular issues (since people who share the same opinions TEND to use the same vocabulary to describe the problems and to cite the same facts). Since this clustering algorithm sorts on rhetorical usage it sorts people based more on how they view the problem than what particular solution they derive.
Two more algorithms can be applied to derive some valuable insight. The first is a timeline analysis - if you notice there’s a group of authors or content outlets whose use of language tends to precede that of the larger group you can correlate the two groups.
Moreover, you can also perform so-called stereotype analysis wherein an author is statistically HIGHLY representative of the group - knowing the variance within the group and the probability of a stereotype can really help identify what is and is not a highly mobilized group of people. This is especially useful for groups whose rhetorical leaders are different from official power structures.
Moreover, correlation is more useful than causation. When you can find individuals that correlate HIGHLY with the larger group you can use them to test stimulus on the larger group, this is idea behind focus groups - testing it on a small group of people reduces the risk of an individual product’s failure. A less-discussed aspect of focus groups is that even when participants reject or dislike the product, the process and procedure of the focus group program has by-far the greatest influence on their opinion of the company.
Taking part in the process improves sentiment MUCH more than receiving a high-quality product We intuitively understand this when we eat burnt homemade cookies or use wobbly shelves we made ourselves.
Everybody understand the value of a mouthpiece but mouthpieces are, in general, a result of the IMPOSED celebrity of a person, not necessarily the fidelity with which their message corresponds with their respective group - see Lou Dobbs on CNN speaking “for Americans” who are of course an entirely homogenous group of hivemind ants. Instead he’s a mouth piece because he uses the sympathies of his audience to insert NEW learned behaviours and mechanisms.
This “though leader finder” machine reverses the process by allowing media organizations to detect and selectively raise the profile of HIGHLY REPRESENTATIVE individual voices.
