Empowering P&G employees to do text analysis

by Scott Reese, Data & Modeling Sciences, Procter & Gamble

In The Sixth Sense, Haley Joel Osment famously says, “I see dead people.” I do not see dead people, but I do see word clouds. Everywhere. Anytime someone asks for help on text, they expect to see a word cloud as the output. Many assume the word cloud is the FINAL output of text, and this is part of the reason I struggle with word clouds.

There are many ways to analyze and explore text (maybe too many!), so why are word clouds so popular? Is it because they are quick, easy, visually impactful, seemingly intuitive and useful? Yes to all, but it’s just a word frequency visualization. There is so much more that can be done!

As one of the two JMP administrators for P&G, I support a variety of teams’ text analysis. My first question to the team is always, “What is the business question you are trying to answer?” The response is often, “We are not yet sure what we want, but we know the answer is in the consumer comments.” (This response tells me that my whole afternoon is about to get blocked off.) My default approach for text exploration and analysis is JMP, which surprises many of my teams. “But I thought JMP was for number crunching,” they say. “We are looking at words.”

To be clear, I have a strong dislike for word clouds – and I know it’s irrational. I use the word cloud to draw the team in, get them hooked and then move on to more sophisticated techniques and approaches.

I like to respond with a three-step process.

Step 1: Engage. Give them what they expect – build a word cloud. I then change the display settings to color the words by a rating and open a filter so they can start to see some comparisons between different products or groups. Less than a minute in, I am no longer asked why I am using JMP. Now the team is asking what other options they can learn so they can go explore on their own. Then the team is off to the races on generating a hypothesis.

Step 2: Educate. Explain the limitations of word clouds, and demonstrate the JMP features that go beyond word counting.

Step 3: Expand. Show the team other options so they can explore on their own.

To be clear, I have a strong dislike for word clouds – and I know it’s irrational. I use the word cloud to draw the team in, get them hooked and then move on to more sophisticated techniques and approaches. Visual exploration of the data is an easier place to start and reduces the barriers to other JMP techniques that produce richer learning and business insights.

At P&G, there is a companywide belief in finding meaningful ways to improve lives. We constantly strive to discover, design and deliver breakthrough products and packages that create consumer delight and sustainable competitive advantage. Word clouds potentially hinder these endeavors, hiding the insights in the consumer comments in a fog. But I’m optimistic that with patience and practice, we will see the light beyond the clouds.

JMP Foreword


Let's stay connected!

You may contact me by email regarding news, events and offers from JMP. I understand I can withdraw my consent at any time.

*
*

JMP Statistical Discovery LLC. Your information will be handled in accordance with our Privacy Statement.