Wow--attending the 5th annual Text Analytics Summit was truly an enriching experience. My compliments to Seth Grimes and the conference organizers for convening such interesting speakers and content. Day 2 started with Sue Feldman and Hadley Reynolds of IDC. Key takeaways from their keynote, Text Analytics Market Report:
- Think about SMBs and the importance of cloud/hosted options.
- Sue referenced research sponsored by EMC with a compelling graphics in Figures 1 and 2. "We're going to hire how many people to manually tag [this content]?!" The need for automated (and unbiased) metadata tagging is huge.
- It continues to be about unified access -- structured and unstructured.
- Rapid areas of growth: social media applications, spam detection, web advertising and search optimization, rich media search, linguistic forensics…
- Insufficient text analytics skills available.
Professor Bing Liu of University of Illinois outlined the challenges of adding structure to unstructured data to do sentiment analysis/opinion mining. This followed by a very interesting end user panel. Chris Bowman, retired Superintendent, Lafourche Parish School Board, coined a new favorite term, "data whisperer." All enjoyed his color commentary. Interesting perspectives were shared by all -- chiefly (in my opinion), the challenge of overcoming people and process issues. Best advice offered: share findings and help make others "look good" not worrying about who gets credit.
My colleague, Manya Mayes, shared "10 transgressions of text" through several cross-industry case studies from her many years of experience in text mining in the SAS and Teragram workshop.
Frank Sun presented four areas of text mining success with big impact at DirectTV. Their effective applications of text mining to the voice of the customer helped them continue (nine years running) to be ranked the nation’s #1 satellite television service, scoring higher for customer satisfaction than all major cable TV companies, in the American Customer Satisfaction Index (ACSI).
Whirlpool shared some learnings from their ongoing journey in applying text analytics, some of which echoed Jet Blue's comments on day 1: it's not always easy to look at negative feedback (let alone to go and seek it in numerous places), but knowledge is always better than ignorance. We applaud this perseverance -- if you don't get balanced feedback, how do you know where to focus your efforts for improvement?
Next, we split into 5 lively discussion groups on these topics:
- Obstacles to implementing text mining
- Frontiers of text analytics
- Customer experience management
- Fraud detection
- Social media analysis
The groups covered some interesting ground and shared summary highlights. Some of the common elements across groups were challenges with people and process, shortage of skills, and issues in getting at the right data quickly and in the most useful form.
Though I missed F2009 (heard it was great!), getting to experience Text Analytics Summit was rewarding on many fronts. It was great to see so much growing interest in the topic and advances in deriving more value from text. Talking to practitioners about their projects, challenges and opportunities is always a treat -- and there were many very interesting conversations. Hearing industry analyst perspectives and research findings -- including The 451, Hurwitz and IDC -- not to mention seeing friends at Data Miners, Tom Davenport and Usama Fayyad, at a well-coordinated event with esteemed colleagues was wonderful.
I will make every endeavor to attend again (happy to report the train home was on time!).
1 Comment
"Chris Bowman, retired Superintendent, Lafourche Parish School Board, coined a new favorite term, 'data whisperer.' All enjoyed his color commentary."
You are most kind! Incidentally, as one “data whisperer” to another, I had a blast. And I have also gotten a lot of mileage out of my comment about how text analytics brings humanity to statistics. My other comment asking, "Would you take a chance on a lottery ticket if I told you there was an 85% of winning," seemed to resonate with people who wonder how you handle the criticism about absolute accuracy. I always thought that 85% trumped ZERO PERCENT. What would you do if a doctor told you had an 85% chance of survival if you got a specific treatment for some dreaded disease? Would you do what he said to do? The answer is unequivocally yes.
This was my fourth trip to that conference and to the users’ panel. I got there years ago when I called to attend, and I told the conference organizer what I did. He thought it was so weird he asked me if I would be on the panel then, and I have been there ever since. I have been to a few education conferences during that four years, and I have never met anybody who analyzed what I analyzed. The interesting part is that I can barely get anyone to talk to me about this or show the least bit of interest. I have often thought that public school systems collect everything and analyze nothing. God knows we need it with the way school systems are pressured today to produce. I also think that the business world is not there yet either except for some firms or industries.
One reason why I like that conference so much is because it is usually populated by extremely smart and passionate people. Though some of it goes over my head, I love to hear about cutting edge stuff. I am amazed at how quickly things change. Last year blogs were hot and now passé because it’s a given that you check them out.
I regret not hearing more about fraud detection. I believe that will grow. There was someone there who works for a legal group learning about that topic. I talked to her on the first day. We talked about ediscovery and Enron. I can only imagine that subprime mortgages, credit default swaps, derivatives, and Wall Street will get the same kind of scrutiny. Several years before this crisis, I read a lengthy piece in the Economist about this. It detailed everything in a frightening prediction that has since come true. If I were in a law firm, text analytics would be my middle name.
Now, let’s move on to sentiment analysis. This has got to grow by leaps and bounds. I did a survey when I was the superintendent about the voice of the employee. One of the things that bothered them was how their checks usually had some type of payroll error. You could also feel the anger and the animosity drip from the page as you read what the respondent had written. Negative words and red highlights dotted the data. Sentiment analysis is not there yet because humor, sarcasm, and irony are hard to identify; however, I learned something about computers over the years. Anything that is impossible is usually done in eighteen months in California. As a novice, generic human being, I find it incredible that text analysis is so far along.
If anyone is even remotely interested in text, this conference is the place to be.