OK, so if you´ve read my previous entry, and have not yet written me off as some kind of anti-technical pseudo-intellectual holistic touchy-feely softie, I´ll now do my best to put send you over the edge. How is it possible that seemingly simple feature of accessibility to and control over the modelling process creates so much value? My reason for saying this is rooted in my experience of being manager of an application support team for a few years a while back. When you are in charge of ongoing support for anything, be it an IT application, hot water boiler, public park, or anything else, you come quickly to understand that the initial of a product or service, while important, is not nearly so as how it develops over time. Taking an IT example, I would much rather start with a mediocre application, but have the ability to adapt and improve it than with a top notch application that is difficult and cumbersome to change. I have seen this effect in many industries and with many products. I know this is not always true, but as a general statement of principle I would suggest that one of the most important, and overlooked, features of any IT application is its flexibility and support for customer customization.
I would also argue that adaptability is especially important in text analysis modelling. Natural Language - the way in which we express ourselves to our friends, customers and vendors is constantly changing. In the past decade we have seen a whole new language develop in the form of text messaging. And, of course, as young people mature, they take the language of their generation to the marketplace, first as customers, then as colleagues, partners and vendors, finally as managers and thought leaders. Adding to this complexity are regional, cultural, age and industry differences in how people express themselves. Thus the language we use to communicate ideas, while based on a shared foundation, is constantly changing. Taken together, I would suggest that any text analysis model is on the road to being obsolete the minute it is deployed, and that a text analysis model is simply a stage in a never ending process, rather than a finished product.
Viewing it in this light, then the ability to quickly adapt and refine a text analysis model is of crucial importance to the customer - and their ability to control that process of great value to them. Going back to my previous example, as a customer, I would very much rather having a text analysis model that only barely does the job but allows me to update and refine it either myself or with people who understand the content, than a model that is initially very accurate but is difficult and expensive to change.
I believe very strongly that text analysis modelling, and ultimately content analytics, will move in this direction. At this point, we are pretty early in this solution/product life cycle. Customers are just now becoming aware of the possibilities this technology offers. They are happy with this first generation of web sentiment analysis that offers them decent initial performance and data, but is out of their control. As we move along the maturity curve, I believe this will change, and change rapidly. Soon enough, our customers will no longer be possible with a black box solution that works - or seems to work. They will want to open up the box themselves, look inside, shake it up, and remake it as they see fit.
Text Analysis - A Never Ending Story
Mark Rice 270001GNUR MRICE@nl.ibm.com 935 Visits