OPINION28 October 2011

What the Foucault? Uses and misuses of social media research


Esomar 3D ventures into the “wild gardens” of social networks and internet forums. Tom Ewing follows.

What better way to shake off yesterday’s post-lunch sleepies than with a great big dose of Michael Foucault? The French post-structuralist doesn’t often feature in research conferences but Vision Critical’s Ray Poynter set out to change that with a high-speed rocket ride through discourse analysis. His talk offered very brief introductions not just to Foucault’s work in this sphere but to conversation analysis, discursive analysis, and recent advances in our understanding of attitudes and memory via social psychology.

Summarising this presentation isn’t easy – the conference Tweetstream was more baffled than usual – but it was enormously stimulating. What I took out of it was an emphasis on the way almost everything said and done online (and offline) is responsive and constructed in the moment: attitudes, common sense and memory are constantly created and re-created, not solid and researchable things. As Mikhail Bakhtin put it: “Everything that has ever been said has been in response to something.”

“Almost everything said and done online (and offline) is responsive and constructed in the moment: attitudes, common sense and memory are constantly created and re-created, not solid and researchable things. This puts text and sentiment analysis, in particular, in a sticky position”

This conclusion puts text and sentiment analysis, in particular, in a sticky position. Poynter had harsh words for aggregation (“just counting words”) and sentiment analysis (“snake oil”) though the breakneck pace of the talk meant no subject was stuck with for long. He rounded off with a wishlist of things discourse analysis could bring to social media – understanding the dynamics of turn-taking on online communities, for instance, or the analogies for conversational pauses and repairs in online conversation.

Gregor Jawecki of Hyve in Germany then presented his overview of a decade in netnography research, focusing on its uses in innovation research. Netnography, according to Jawecki, is based in passive observation and analysis of consumers’ natural behaviour online, particularly in forums. And there’s a lot of behaviour out there: he gave an example of a chainsaw forum in which its 6,722 members contributed north of 650,000 posts.

With this as its field of study, the potential of netnography is obvious: Jawecki described it as a great tool for scoping a new field, extracting the bulk of people’s knowledge in a particular area by seeing what its communities of interest say about it. He shared a case study of a deodorant study where netnographic work established that deodorant staining was a major consumer concern, and was used to create a “stain manual” of the types of stains, as defined by consumers themselves – as well as the solutions they found to the problem. This work fed directly into product design. Jawecki was persuasive and positive about netnography’s benefits, though he acknowledged that greater integration with traditional quant and qual would be its route to fuller acceptance.

Finally, Annie Pettit of Research Now’s “I’ll Tell You What You Want, What You Really Really Want” spiced up our lives with a dose of sarcasm, by setting herself the mama of all research reporting briefs: proving consumers like bank fees more than cookies. How did she manage this? She cheated. In her example, a study on cookies should distinguish between the tasty kind and the computer kind, but with the wrong parameters the two could easily become one. Social media data can be manipulated, and experimenter expectancy effects might mean too much positive or negative sentiment.

Pettit wants social media researchers to stop right now with their poor sampling, bad keyword detection and sloppy sentiment scoring. The message? Let professionals lead the way, and watch out for social media monitoring wannabes, or be prepared to say goodbye to consistency.

Editor’s note: Tom Ewing clearly knows his Spice Girl tracks – the last two paragraphs are littered with references to the group’s biggest hits. There’s a £10 iTunes voucher in it for whoever spots them all. Just tweet us the following message: “@researchlive – I found XX Spice Girls references in @tomewing’s #eso3D review http://bit.ly/tfBH7i” – XX being the number of references. If more than one person finds them all, a winner will be drawn at random. There can only be one. Editor’s decision is final. Winner will be announced at 10am on 31 October.

Update: The competition is now closed and we have a winner – @words_said on Twitter found all eight song titles, which are (in order of appearance): Spice Up Your Life, Mama, 2 Become 1, Too Much, Stop, Let Love Lead The Way, Wannabe and Goodbye. Congratulations.

Tom Ewing is digital culture officer at BrainJuicer


13 years ago

Great to see a Bakhtinian approach being discussed so prominently in a Market Research setting. Re Discourse and Conversation Analysis - Aggregation is a very limited approach to text analysis and Wordles are the most visible aspect of this simple approach's misuse - they've recently received a lot of criticism! Sentiment Analysis is of course not perfect (and I'm not sure that's the standard it should be held to - if it's useful then that's a good starting point) but building in other elements from across linguistic theories creates many opportunities to better understand what consumers are saying.

Like Report

13 years ago

Thanks for the summary, Tom. I have put together a few slides on the Nivea case mentioned above: http://value-co-creation.blogspot.com/2011/03/from-insights-to-market-launch-niveas.html

Like Report