Recently, there has been much discussion that social media may drive down the use of surveys and focus groups. Some have claimed that surveys could become obsolete in the next 20 years. That’s an interesting concept, but some of the talk may be hyperbole designed to attract an audience. One thing is certain, though: Social Media cannot be ignored.
Platforms like FaceBook, Twitter, Company Websites, Blogs and Forums all provide a rich, valuable source of customer feedback. Unlike more traditional MR data sources, the Social Media participants are running the show. They decide what’s important enough to them to talk about, to respond to, and to debate.
Dipping into this sea of commentary is the digital equivalent of eavesdropping at a cocktail party – you’re almost certain to hear something juicy, but you better have a sense of the context before you go repeating it!
And therein is the challenge: How does one get the most out of the Social Media data stream? How can we separate the signal from the noise? How do we define and identify a trend? What can we assume, or not assume, about the respondent demographics to know if they speak for our market, or for a segment thereof?
This Text Analytics challenge is the focus of many in the MR community. It’s an exciting and formidable frontier, because unlike the numbers of mathematics and statistics, all languages do not obey the same rules of grammar and syntax that we must use to parse out the meanings from the prose. Researchers are already using Text Analytics on SM sources with some success in discovering evolving trends, finding potential PR issues and collecting some crowdsourced opinions. Some are basing their work on an extension of the tools used for Open Ended Coding, while others are developing entirely new approaches.
So far, the toolbox seems to be incomplete. We’re not at the stage where we can send in a bot and pull out all of the goodies. Some of the hurdles are large, but the minds working on this are intelligent and dedicated. It’s a big task: Any tool must have at least language-specific versions, with their attendant cost and complexity. And, it must go far beyond word counts – what does the number of mentions matter if the context is unknown? Even “simple” sentiment detection is not so easy; English alone has dozens of different constructions for expressing positives and negatives, some of which depend on word order alone. Whoever gets this right, or close to right, will get us closer to true AI than MR has ever been.
The survey isn’t going away, though. While its use will reduced as our ability to mine the Social Media data grows, we will still find many applications where the traditional survey remains the best available option.