Intelligent Enterprise

As Curt Monash reports in his Text Technologies blog, Voice of the Customer was a central theme at this year’s Text Analytics Summit. The aim is to stay on top of reputation, quality, and product-design issues by crunching blog- and message-board text, call-center notes and e-mail, and free-text survey responses. (Some vendors call these activities "Enterprise Feedback Management.") Yet VOC and the analytical approach it typifies are only half the overall text-analytics picture. Text analytics still delivers very high value in traditional, non-VOC application domains such as life sciences and intelligence, areas where vendors still derive the major part of their revenues.

Work in these latter areas is investigative. Users hope to spot needles in haystacks (without even always being sure what those metaphorical needles look like) rather than the broad, statistically characterized trends that the VOCers are seeking. As shown by the healthy turn-out at last week’s Basis Technologies government users conference – I would guess that Inxight Federal Systems attracted similar numbers to their similar event this week, which I couldn’t make – continued interest in investigative text analytics is out there even if the weight of attention at the Boston Text Analytics Summit was on the VOC side.

Yet we did get plenty of traditionalists, for instance, Breck Baldwin of Alias-I, who gave a great talk on "99% Fact Extraction from Text," a must if you’re trying to thwart terrorists and desirable in many other applications even if you can get by with much lower accuracy when your concern is consumer-goods trends. The presence of both those traditionalists and of VOCers – and of plenty of veterans who understand both needs – led to recurring discussion of accuracy requirements: how good do you really need to be?

The reality is that VOCers are as interested in accuracy as traditionalists, just in a different way. Where the investigative types want high f-scores – values computed as the harmonic mean of weighted precision (is what you identified correct?) and recall (did you get everything?) figures – the VOCers similarly prize precise and exhaustive identification of significant trends. That is, accurate trending can paint as true a picture, albeit from a different perspective, as approaches that retrieve documents and extract information with high precision and recall.

So Voice of the Customer: welcome to the text-analytics fold. The field can and will accommodate both trend-spotters and investigators with flexibility to suit many sorts of end-user goal.

Seth Grimes is president of Alta Plana Corporation, a Washington DC consultancy that specializes in BI and advanced analytics.

»  Digg
»  Slashdot