Intelligent Enterprise

Damn trademarks. I'm slated to speak on predictive text analytics at October's Predictive Analytics World conference near Washington DC. Release of the PAW agenda elicited a twitter comment from my friend Olivier Jouve, "Seth, glad to see you using 'Predictive Text Analytics' - expression that SPSS and I crafted in 2003!" (I’m at @sethgrimes on twitter by the way.) Olivier is SPSS vice president for corporate development. He and SPSS do deserve credit for promoting wide commercial deployment of text technologies that had previously been accessible only to researchers. I only regret that before titling my PAW talk, I hadn't realized that SPSS had trademarked "predictive text analytics," turning a term that deserves wide business application into SPSS property. Had I known of the trademark, I would have chosen a different title. While SPSS hasn't trademarked "predictive analytics," a more general term that has been in use for years, I'm impressed with the company's ability to execute on a broader vision to similarly own that field.

SPSS Senior Vice President Colin Shearer sketched out the company's Predictive Enterprise vision for me over lunch in May, 2005... literally, and for some reason I taped his sketch to my office wall. (I have lots of stuff posted including, given my interest in visualization, a variety maps and infographics.) It includes predictive analytics, albeit on equal footing with deployment and decision optimization. At that point, in 2005, the company had already been talking about predictive analytics for some time, trotting out a definition by Gareth Herschel of Gartner, "Predictive analysis helps connect data to effective action by drawing reliable conclusions about current conditions and future events," and applying tag line, "Understand, Predict, Act." SPSS's latest release, PASW (Predictive Analytics Software) Statistics 18, announced on July 14 and slated for mid-August availability, is a next realization step.

SPSS text technologies belong to different software stack, however, with a different software update cycle. The linguistic elements trace their origins to LexiQuest technology acquired by SPSS in 2002. Olivier’s tweet pointed readers to a 2003 write-up by Philip Howard of Bloor Research that defined predictive text analytics essentially as information extraction from textual sources for joint analysis with numerical data. That genre of unified analytics has indeed proven its value in applications that range from customer-experience management to fraud and risk modeling, where one might link text-extracted information with structured transactional, demographic, and behavioral data. Contrary to Bloor's 2003 view, I do include clustering, categorization, and classification of text and text-extracted information -- of documents (HTML, PDF & DOC text as well as audio & video clips) and of the "features" they contain -- as a predictive function. This variety of prediction doesn't involve forecasting time based variables; rather it involves automated decisions about the nature and meaning of often-ambiguous data objects based on the analyst's inferred purposes and intent.

In my PAW abstract, I describe predictive text analytics, naturally enough, as predictive analytics applied to textual material. This short description doesn't capture an essential point, that text analytics brings semantics -- discovered, contextually relevant meaning -- to the practice of predictive analytics. There are some great text analytics solutions out there from SPSS and other companies and some really illuminating use cases and user stories. Some are extending predictive capabilities from entities and facts to concepts, events, and sentiment, making for even richer analytics. There's a lot to learn here and a lot to talk about. Stay tuned!

E-MAIL | FOLLOW US