Zeitgeist refers to something that any of us can relate to: social sentiment, in other words, the spirit of the times (these we live in or previous eras). It’s also one of the most pretentious words in the English language (the fact that it is obviously borrowed from another language adds to the pretentious factor). Every self-important commentator feels that they have their finger on the zeitgeist of something or another. At heart, it’s simply their perspective, which may be entirely impressionistic and intuitive, though some of the best analysts offer compelling factual details to bolster their assertions.
However, social sentiment analysis, as we practice it with big-data marketing analytics tools, isn’t quite so grandiose. It’s not zeitgeist surveillance, in the classic sense. Instead, social sentiment analytics is a highly focused species of “zeitgeist.” It can help your marketing analysts better assess how specific customer segments are feeling right now about some specific topic, such as your company, brand, or products. To a growing extent, marketing professionals in all industries are relying on social sentiment analytics technologies to take impressionistic judgments out of the equation and thereby (it is hoped) allow them to base their go-to-market strategies on hard facts.
Fat chance of that. What many marketing professionals don’t fully realize is that you can’t do social sentiment analysis on auto-pilot. No matter how sophisticated your tools, you will always need human interpretation to contextualize and identify the true meaning of all the verbal “facts” you harvest from social networks and other sources.
To my great amusement, a recent article illustrates this critical point. The piece discusses the most common analytic approach–sentiment analysis by word count–and highlights its chief vulnerability. It describes a big-data project that attempted use feeds from Twitter and other social media to predict the U.S. unemployment rate. The researchers looked for correlations between monthly usage volumes of specific keywords and monthly unemployment rates. During one month, they noticed a sudden spike in mentions of one keyword–”jobs”–and thought this was significant, not realizing that this might have had something to do with the death that month of one of the co-founders of Apple Computer.
This sort of interpretive error is common in sentiment analysis by word count, and, as the Harvard professor cited in the article notes, it’s even more prevalent when you attempt to make more wide-ranging sentiment analyses that consider many keywords. If you go that route, and don’t have human analysts providing a reality check by actually reading a random sample of the tweets that contain these keywords, you might make increasingly absurd “zeitgeist” pronouncements based on correlation spikes among homonyms (a la the “jobs” example) and other verbal flukes.
Interestingly, the article takes it one step further to point out how even human “experts” can totally misinterpret the source data if they are trying to interpret verbal nuances among unfamiliar populations. It cites a study where American doctors, unfamiliar with tropical diseases and trying to conduct autopsies by interviewing the families of the deceased in Tanzania, might undercount the incidence of malaria due to their failure to recognize its symptoms.
One could bring that last insight back to the world of social sentiment analysis. If you’re analyzing customer sentiment among a demographic to which you personally don’t belong (i.e., another age, racial, ethnic, religious, regional, socio-economic, or other group), you’re quite likely to misinterpret what you read in their social-media feeds. In those cases, you’re like the many anthropologists who’ve been misled or outright deceived by the peoples they studied. You’re just not hip to when they’re pulling the wool over your eyes–or, even if the people are on the straight with you–you might not recognize how clueless you are in interpreting things about other cultures that are right in front of your nose.
If your marketing professionals aren’t using big-data analytics to tune their personal social-listening aptitudes, you’ll be squandering your investment in the technology. Social sentiment analysis is only valuable if it empowers marketing analysts to make more valid, precise, nuanced, and contextualized interpretations.