Since the rise of e-commerce, we've been told to focus on what customers do in their dealings with us. But in the past few years, as we have become increasingly media chatty — social and otherwise — we are “rediscovering” the value of knowing how people feel about us, in the form of sentiment analysis or opinion mining as it is also called.

Mind you, this isn’t new. If you sell stuff, it has always made sense to be interested in how your prospective and actual customers feel about their interactions with you. In fact, formal content analysis has been around at least as far back as the late 1920s, largely through face to face interactions or via phone, mail, surveys and the like.

What’s different today is that the academic community, beginning in earnest roughly 15 years ago, discovered that they could use automated statistical techniques to find embedded meanings in masses of textual materials, both authored and related to vendor-customer interactions. Burgeoning computer power made it feasible and the rise of big data provided the raw material leading to automated sentiment analysis.

3 Schools of Thought

As with every new application of technology, the new statistical approaches to sentiment analysis have generated a range of enthusiastic adherents, skeptics and downright detractors. So what are we to do if we want to find out what people think about us? It’s confusing at best.

Search for Automated Sentiment Analysis, for example, and you confront three distinct schools of thought, all seeking to influence you and what you do next:

  1. It’s the greatest thing since sliced bread, a new horizon in dealing with market response to your products, providing insights that will materially improve your results .... You hear this take mostly from firms that want to sell you automated sentiment analysis systems or tools.
  2. It’s a blunt instrument and, while improving, has major limitations and should not be used without significant human backup .... You'll hear this from firms offering you mostly human alternatives to automated sentiment analysis.
  3. It’s a fascinating new area of research in statistical analysis, providing a fertile path to better use of the inherent power of big data and analytics to monitor opinions and references in mass postings on social media and other forms of mass communications …. This point of view comes from an academic community focused on the elegance of the process and viewing any statistically significant improvement, however modest, as a major victory.

So how are we to find out what people think of us? A few basics come to mind:

  1. Human analysis, if you can do it, is still generally regarded as the most accurate and revealing way to measure your audience’s feelings about you. While automated sentiment analysis is much in the news, there are still powerful tools available to measure sentiment. On-line post transaction surveys, follow up calls, email surveys and the like can put you in touch with customers to determine how you are doing with them.
  2. Simple is often better, so if you can measure your customers’ sentiment without resorting to the complexities of mass statistical analysis, you may end up with a better result at considerably lower cost and risk. Indeed, because automated sentiment analysis is still moving from controlled laboratory experiment to commercial product, the jury is out on how successful the transition is and how much you can trust the commercially obtained results.

Where Automated Analysis Makes Sense

Where the sample is large, automated analysis can be good at spotting extreme trends at the margins of negative and positive. It can also be useful to narrow the sample to something that can be manually evaluated, or to spot extreme good or bad trends that don’t need further evaluation.

Most automated scales in commercial use today confine themselves to rating text material on a simple one-dimensional scale: positive-neutral-negative, using a number of different statistical approaches applied to text entities from entire documents to paragraphs, single sentences and even clauses. If you’re looking for a more in-depth picture, you may have to resort to human analysis.

While there are commercial sentiment analysis products that attempt to measure attitudes at a much more detailed level, attempting to mine the text for nuances of attitude, correcting for sarcasm, irony, hyperbole and other textual subtleties remains a challenge. These products can provide more nuanced results, but quickly lose clarity and accuracy as they attempt to describe more detailed variables.

Learning Opportunities

Keep Your Eye on the Goal

For those in commerce, the goal is to learn what you are doing well, what is hurting you and what changes are most likely to fix the problems. While statistical trends in a massive data sample can sometimes be useful, each member of that sample is and acts like a real person in making buying decisions, and homogenizing them into a statistical measure of attitude, however elegantly it is done, may miss key details that could direct your product and marketing efforts.

Another problem with mass statistical analysis is that customers who have negative feelings about their interaction with you or your e-commerce applications often decline to leave reviews, they just never come back. This can skew the evaluation toward the positive and mask serious challenges you should know about. Selected outreach to one-time customers, if carefully done, can help to uncover some of what might have been missed in the automated analytical process.

On the other side of the scale is the fact that those millions of people on social media often find themselves using gratuitous hyperbole in their posts, potentially skewing the classification of their input toward more negative evaluations than they actually feel or will act on. And those 140-charactrer tweets make nuance virtually impossible.

If you still feel driven to automate your sentiment analysis on a mass basis, a hybrid approach using both automated and human analysis may insulate you from the shortcomings of either. The key to success is to look closely at the results. If they don’t make sense to you, decisions made based on them probably won’t either.

Whatever Path You Choose, It Won't Be Dull

Even with the ambiguities surrounding what techniques are most effective, the rise of sentiment analysis is valuable if only for its clarion call to care about what your market thinks about you, however you put that care into action. Big data analytics, for example, told J. C. Penney that their customers bought more based on lower regular prices, so they lowered them and dispensed with their traditional sales and store brands. Sentiment Analysis would have told them that those customers valued sales and would go elsewhere if they disappeared, saving them billions of dollars and a large part of their market, which now may never return.

In a commercial world that can change overnight, it’s critically important that you stay sensitive to how you are viewed in your target market. If you go about it carefully — and skeptically — you can create an environment to learn more about who is saying what to whom about you and what it means for your future.

Creative Commons Creative Commons Attribution 2.0 Generic License Title image by  Tobias Zils 

fa-solid fa-hand-paper Learn how you can join our contributor community.