Text Analytics 2008 Redux

You’ve had Nick’s take, now here’s mine, with a little overlap – great minds think alike, right? 😉 We were not expecting the 40 attendees for the pre-conference workshops during prime Sunday TV viewing time. Seth Grimes laid out “Text Analytics for Dummies,” while Nick gave a market overview. But the attendance (and the long Q&A sessions) were good indicators of user enthusiasm and the desire for real, practicable advice about the field.

Some of the other memorable moments:

  • Best of the vendor panel: Seth Grimes’s challenge to say something nice about a fellow vendor’s offerings. And the vendors’ response to an audience question about incorporating UIMA, which was uniformly that it wasn’t necessary or in demand.
  • The Facebook presentation on trend-tracking through users’ “Wall” posts was brought back for an encore by popular demand. The crowd in my session was a little confrontational about the amount of analysis being done on the available information (never enough!), but as far as quick and dirty zeitgeist goes, it was unbeatable, and a lot of fun.
  • The Clarabridge 1-hour deployment was good sport, with at least one customer’s testimony that once the system is learned, it can actually be configured with speed approaching that of CTO Justin Langseth. You have to hand it to Clarabridge: they make it look easy.

Some thoughts on the users’ takes:

  • In presentations and in private chats, frequently recurring themes among vendors was eDiscovery and social media – some of the drivers for the market. The user questions I heard were mostly about sentiment analysis, deployment time and ROI. Specifically, information on how to judge all of the offerings – is sentiment analysis accurate enough? What is the expected deployment time, what is the ROI?
  • Precision and recall went back and forth again, but the hard truth is that the edge depends on the application. For patents or PubMed searches or eDiscovery, you need recall. For other applications, precision is paramount. Some users I spoke with mistook this as a lack of accuracy – it’s more of a sliding scale of usefulness.
  • Accuracy was a recurring issue, both because text analytics is an emerging technology, and, of course, text is messy and imprecise. Partly it’s a matter of maturation. But the “fast / cheap / or good – pick any two” truism about software development is equally true here. Even with built in taxonomies and dictionaries or domain-specific knowledge, any text analytics software needs configuration to increase accuracy for its application and user, which takes time.
  • “Win fast and win often” – great words from Tony Bodoh of Gaylord Hotels, on the user panel. Because of the financial investment, the fact that text analysis software can automate (obsolete) some employee work, the time it takes to configure, and general resistance to change, it is important to gain both executive and user buy-in early in the process. Chris Jones of Intuit echoed the sentiment, adding that it’s not advisable to go after your largest (and most time-consuming) problem first – come up with a number of smaller successes to prove the concept to users and higher-ups. Incidentally, both of these are Clarabridge users.
  • Jones also noted that one of his “lessons learned” was to avoid over-configuring or too much tinkering with the analytics. He advised after a prudent amount of configuration to treat it more or less like a black box, and not worry about what is going on under the hood, just let it do its job and leave it to the professionals.
  • Some more wisdom from the user panel: you can’t go into a text analytics deployment expecting quantifiable ROI. “You don’t know what you don’t know” – which is what the tool is there to solve. In many cases, the real potential isn’t obvious until you can see how it works with your business. At that point it’s possible to come up with applications that not even its creators could have thought up.
  • Lastly (and this is not a new sentiment, but it meant more coming from school Superintendent Chris Bowman, who looked like he had my parents on speed-dial): the text analytics field is emerging, and will become integrated with larger applications. This will eventually render a conference like this obsolete, but it also means a great chance to get a leg up as an early adopter.

Looking forward to next year!