Text Analysis for Transactional Surveys
Posted by Jeffrey Henning on Tue, Jan 05, 2010
Recently in a Vovici research webinar, Catherine H. van Zuylen (@Catherine_VZ), VP of Product Marketing for Attensity Americas, co-presented "Social Media & Surveys: Mining Sentiment & Attitudes to Understand the New Breed of Customer". Catherine said that many people ask her if they really need text analysis, or if they can get by just reading verbatim responses or using keyword counts.
The problem with reading all the verbatim responses to open-ended questions is that, while people start with good intentions, they usually don't get past the first few hundred. Sometimes they just end up doing a word cloud or doing a keyword count. This may be fine for a survey project, but now think about a survey process, where you are running a satisfaction survey every day, collecting new verbatim responses hourly. That's a lot of work. Now broaden it to read all the comments about your brand that are made on web sites each day: if you have a popular brand, you'd need to add staff to do nothing else full time.
So for transactional surveys, you need some text analysis. Keyword counts are a simple analytical technique, but they ignore sentence structure. For instance, analyzing hotel reviews and tallying "room" and "smell" produce false positives like "Although it was a smoking room, I could not smell anything - it seemed to be well-ventilated." A keyword count would index that as a problem, when in fact it is a compliment. Text analysis, in contrast, parses sentences:
Keywords can alert you to an issue: for instance, lots of rental-car customers are mentioning "cupholders". Advanced text analysis can help you connect the dots, alerting you to the fact that the cupholders are too small or that there aren't enough of them.
Text analysis can also help you route information effectively. The sentence "I will close my account if no one calls me back" produces, among other structures, "I - close[if/then][intent] - my account". With this intent discovered, an alert can be routed to the appropriate customer service representative to take action before this customer takes their business elsewhere.
You can certainly live without automated text analysis for one-time survey projects with a few hundred responses. For transactional surveys, though, advanced text analysis helps you take advantage of the hard work that you've done collecting feedback (whether from survey data or from mining comments about you on the Web) and rapidly and efficiently take action on that feedback. Watch the webinar to learn more.