Is Text Analytics Ready for Prime Time?


      Bookmark and Share

I have previously reviewed basic business practices required to anticipate and manage inevitable social media crises affecting brands. Even the best business practices can only mitigate social public relations (PR) disasters. And it is expensive, and of dubious value, to do as the Department of State is reported recently to have done: spend $630,000 on Facebook likes. That's probably, at best, a one-off solution. 

Can text analytics tools (including sentiment analysis and predictive analytics) provide an edge to anticipating or even understanding Big Social Data? Vendors are hopping on the bandwagon. But what about the state of these tools themselves? I polled four vendors, large and small, to get their views about the current state of these tools. The vendors are Adobe, Lexalytics, Semantria, and Recorded Future.

The "Internet of Things" is leading to the datafication of everything. This datafication extends from highly structured (database) information to loosely structured text. Data by itself, though, is useful only when we can analyze and act on it, even predicting crises before they occur. Matt Kodama, VP of Recorded Future, defines predictive analytics as "using statistical methods on datasets, to either test a hypothesis that some specific pattern or correlation is strong enough to have predictive power ... or to generate correlations for experts to consider." Kodama believes that text analytics and sentiment analysis complement predictive analytics by working the document portion of those datasets.

How effective can sentiment analysis be at detecting a common problem, such as "tag jacking" of Twitter hashtags, which is using sarcasm or worse to damage brands? Kodama believes the human element is always needed, and the tools today are generally immature "because people are so good at inventing and comprehending creative use of language." Adobe's John Bates, product manager for predictive marketing, is more sanguine: "We are quickly seeing a transition in varying degrees towards the next stages of analytical maturity in text diagnostic analytics (text mining) and predictive text analytics." Bates points to Adobe Social's predictive social publishing capabilities announced at this spring's Europe, Middle East, and Africa (EMEA) Adobe Summit.

Seth Redmore, VP of marketing and product management at Lexalytics, is disarmingly candid, saying that sentiment analysis' weakness is that it requires understanding humor or sarcasm, for which there is no good solution. He insists that the human side is always important: "It's important to be a subject matter expert and understand the limits of text analysis and the tradeoffs." Finally, Semantria's marketing and analytics manager, Scott Van Boeyen, claims that up to 50% of predictive analytics and business intelligence markets rely on text analytics. He believes that will rise to 100% in 2 years.

Let's get back to the Department of State's purchase of Facebook likes. The U.S. and other English-speaking countries make up less than one-third of all internet usage, but even with Facebook-the largest social media site-buying those likes in foreign language settings is complex, and it indicates yet another weakness of analytics tools. Those tools must also be multilingual, and they don't always meet that need. All this leads me to two general observations.

First, text and predictive analytics are really difficult, but, with the increasing volume of information, it is critical to kick the tires and get familiar with these tools. To use Gartner, Inc.'s terminology, we are somewhere between the "Technology Trigger" and "Peak of Inflated Expectations" points in this cycle of technology. Ask for demos. Some companies, such as Semantria, even have an Excel-based sentiment analysis tool they will let you try out.

Second, the volume, variety, and velocity of Big Data, within and outside every organization, means these tools touch many organizational areas and interests. Although you always want to start small and scale up your solutions, it is important to ask what other organizational tools or interests already in place might be affected. Think creatively.

Do text analytics tools work with other solutions you may already have in place? Why not just use the in-house search systems you already have invested in and support, instead of buying new sentiment assessment systems? I'll explore these topics in future columns.