Get the early bird discount when you register now for KMWorld 2017 in Washington DC

Things to keep in mind when visualizing text

   Bookmark and Share

So you’re had success in building a text analytics model.

Now you want to visualize the results and take action. What are some things to keep in mind?

Let’s start from the beginning...

A text analytics model will generate new variables (such as categories, sentiment scores, data-driven topics, entities, etc.). These variables become even more meaningful when they are used to model customer attrition, identify cross-sell opportunities, monitor critical category trends, and uncover relationships within the data that help to enhance sales or reduce costs. A successful text analytics process will output relevant variables, structured in a meaningful way for visualization.

Start with a strong data foundation

What is the best way to structure the data for visualization? The answer depends on your visualization or modeling requirements. For example, topics that are derived from the text can be trended over time (to detect emerging trends) or used for color coding or sizing. To accomplish this, these variables need to be numeric/continuous. What if you want to create hierarchies, network graphs or tree maps? In this case, the data-driven topics should be categorical.

What should be reported.

What visualizations will answer your questions in the most meaningful way? Hierarchies are a great way to drilldown into categories and sub-categories to uncover root cause. Interactivity across graphs enables users to see the relationships between variables, which isn’t possible with static visuals.

1.) Answers lead to more questions. In most of the engagements I work on, organizations are trying to understand their customers be leveraging text analytics. I've sat in meetings with executives where we present the results and they get all excited and start asking more questions, and deeper very good questions at that.

The very nature of conversation (especially good conversion) is that it leads to questions. Meaningful answers to these questions typically spurs deeper curiosity and even more questions. Text and online conversations are no different.

Enough is enough.

What is the right balance of depth and accuracy required for your business so necessary and how to do your best cut through the noise to get the "essential data-driven answer(s)" that allow you to address your business requirements?

Integration of structured and unstructured analysis

I've seen a wide variety of use cases and requirements around text analytics, and one common theme in most of them is that text is used to facility a larger business requirement. Maybe the text variables are used to enhance a predictive model (in the case of net promoter scores). Text analytics can also be used to categorize call center notes, warranty information, complains, and social media. These categories are used to monitor key issues that can be used for training, customer retention or upsell opportunities. The real value of these new categories, topics, and sentiment scores are uncovered when pair with structured data such as demographic information.

How do you visualize your textual data? Is your goal to explore what is buried within he text and uncover relations, to create executive reports, create social media monitoring tools?

Search KMWorld