Text Analytics: Summarizing the #TMRE Hashtag traffic

The TMRE 2011 conference wrapped up late Wednesday afternoon.  Several attendees were actively tweeting to the #TMRE hashtag which many of us followed during the event.

Last evening I sat down with @dwiggen’s twapperkeeper archive to pull an RSS feed of the #TMRE for analysis.  The feed as of last evening covered tweets to #TMRE from 11am Monday through 11pm Wednesday (event time).  In total, 2710 tweets were captured; 513 Monday, 1,340 Tuesday and 817 tweets on Wednesday.

As shown on the accompanying graphic,  the popular “tweeting times” were the 9am and 11am hours on Tuesday and Wednesday in sync with the excitement surrounding the keynote sessions.  Peak tweeting volume occurring at 11am on Tuesday.

Sure, there’s a tag cloud …
The tag cloud graphic below was created using Wordle after parsing the feed to remove sender, hashtags, URLs, and stop words.  Retweets were included within the analysis in an attempt to lend additional weighting to the tags.  Just over 36% of total tweets were RTs.

 

“Most Prolific Tweeter” goes to …

Excluding event host TMRE, InsightsGal took the honors and was also retweeted most frequently in total, but notably VirtualMR was retweeted more frequently as a percentage of VirtualMR’s total tweets.

The two most frequently occurring RTs shared the concept of thoughtful interpretation of your survey research analysis:

  • Outsource process, but never outsource thinking. -Stan Sthanunathan of Coke
  • You can’t have breakthrough insights with people who all think the exact same way.

Taking the analysis a step further, we used an algorithm to select a small subset of tweets that are representative of the entire #TMRE tweet stream.  The algorithm scores tweets based upon their weighted tag values after adjusting for tweet length.  This resultant “tweet brief” provides a quick flavoring of the prominent themes within the overall #TMRE tweet stream:

  • rachel_bell44: If you don’t like change you’ll like irrelevance a whole lot less! – Heiko Schafer Henkel/Dial Corp
  • eswayne: #TMRE a lot of people ascribe Coca-Cola’s success to a TV spot, but the real power is the communities of people surrounding the brand.
  • MattIIRUSA: RT @Ali_Saland: Ask yourself, how have u made it simple 4 the brain 2 process? What have u done 2 make the consumer feel better by buying ur product
  • mg_hoban: TMRE=Take More Risks, Everyone! RT “@LoveStats: TMRE=t-tests manovas regressions experiments”
  • CandiceSeiger: RT @klonnie: #TMRE social media Anthony Barton@Intel small amount of research paranoia over SM/time to insights is far shorter with SM

I hope you enjoyed this summary.  If you attended the conference this week, I would expect that you’ll see familiar concepts captured within the tag cloud and representative tweet content.

If you’d like to see the data cut another way, I’d encourage your feedback, or contact me to discuss how we could apply this form of discovery to the results of your next survey project.

Survey data analysis – Text Analytics for Net Promoter Surveys

As a consumer insights manager at a national retailer, you are responsible for understanding the wants and needs of your customers.  You’ve followed the sage advice of the Net Promoter Score (NPS) experts, and you’ve implemented an online NPS survey capturing “Would recommend..” that allows customers to leave open-ended comments.

In our prior posts we’d discussed how to join internal “structured” data such as retail location, transaction date/time, payment type, check amount, etc. to your survey results, segmenting the NPS scores to identify  the customer segments requiring the most attention.  But how do you make sense of a 100,000+ open-ended customer comments?  

You can segment your open-ended or “unstructured” customer comments using your “structured” data in the same manner as your NPS scores.  For example, we could focus our review of customer comments to the customers paying with a credit card within the Eastern region reporting an NPS score of 50 or less. Unfortunately, national retailers will likely find that they still have hundreds to thousands of comments to review even within very focused segments.

We can use text analytics to help make sense of survey comments without having to read every individual comment. Term frequency (TF), or the percentage in which a selected term occurs, helps gain an understanding of the important concepts underlying the comments, just like you would use the mean of a series of numbers to gain an understanding of a numerical series.  Term frequencies can be conveniently analyzed using a selectable tag cloud graphic as we’d illustrated in a previous post.

When TF analysis is combined with segmenting the comments using our “structured” data segments, we can compare the term frequencies between various segments to gain an understanding of the relative importance of key concepts within each segment.

We can further refine term frequency analysis by normalizing our TF terms using inverted document frequency (IDF) analysis. TF-IDF analysis goes a step beyond TF analysis by considering both the frequency of occurrence of a term with a individual comment (TF) as well as term’s frequency of occurrence within either all comments or a segment of comments.

We can then use TF-IDF analysis as the starting point for determining the relative similarity of comments. This additional step allows us to call out a small set of comments, for example, the top 20 comments, that are the most representative of the larger group. These top 20 comments serve as an “executive summary” of the overall set of comments, summarizing the key topics and issues without the need to individually review the entire series of comments.

We would welcome the opportunity to help you to reveal the hidden insights within your NPS survey results.  Please follow us on Twitter http://www.twitter.com/mTAB to receive updates to our ongoing survey analysis series as well as more information pertaining to our mTAB survey analysis service.

Enhancing Your Net Promoter Score Survey Analysis

Imagine that you are a retailer tracking your net promoter score (NPS).  Your NPS survey is administered by way of an invitation printed on your in-store POS receipt.

The point of tracking NPS is to identify how to make improvements in your service and products.  Changes in your process can then be validated by continual tracking of the NPS metric.

So how can we best identify opportunities for improvement from the NPS tracking graphic below?

We can simply observe that the NPS score has gone up or down relative to our last observation.  Assuming at least one open-ended question in addition to our net promoter survey question, we can review the open-ended or “unstructured” data from customers providing low (or high) NPS scores.

Text analytics tools, for example tools providing classification and sentiment analysis of unstructured survey responses, along with search term and tag cloud drill down tools, can significantly enhance the understanding of the unstructured survey responses.

Additional structured data, coupled with the appropriate tools for the analysis of structured and unstructured survey questions, will significantly enhance our analysis and thereby greatly increase the value of our net promoter program.

POS receipts typically include a transaction identifier that can be used to link a wealth of structured data to the survey results.  Listed below is a partial listing of possibilities:

  • Store Location
  • Transaction date/time
  • POS terminal and sales associate
  • Inventory of items purchased
  • Items purchased on sale or promotional items
  • Payment type
  • Total amount of purchase (total check)
  • (assuming you have a frequent buyer program) buyer profile

Without adding questions to our NPS survey, we can now break down scores by region, district, zone and store, by day-part, by payment type, by check amount, by store department and potentially by frequent vs. infrequent shoppers.

If you would like to review your NPS program with industry experts  please sign up for a web meeting or simply visit our web-site to learn more about our survey data basing and analysis services.