When Data Has Lost That Lovin’ Feeling

By: Lisa Sigler

February 24, 2015

Tags:
social listening
social media

Figuring out how someone else feels about you is tricky. It’s a problem at the heart of countless pop songs and romantic comedies.  From a business perspective, it helps you to make forecasts, plan campaigns, and gauge customer loyalty – but only if you have a way to figure it out.

Sentiment analysis is the examination of data to determine if the expression is positive, negative, or neutral, and to what degree. It gives you a way to use standardized, automated methodology to understand those fickle, illogical things called human emotions. It’s a discipline that is applied to unstructured data – basically, to text. This text can be captured in an email, the “additional comments” section of a survey, voice recordings of customer interactions, notes in a customer support log, a post on a customer review site, in social media, and dozens of other places.

(This is distinct from structured data, which is the kind of information that is generally found in a survey: name, location, age, and 3 out of 5 stars, for example.  You can’t get a lot of emotional information from looking at structured data – names, purchase history, even rankings from “1 for love to 10 for hate” only give you so much.)

Combining both structured and unstructured data analysis is obviously important if you want to get to the heart of the customer. However, there is even a lot of variability in sentiment from different types of unstructured data. For example, social data tends to skew positive – people don’t want to be perceived as negative on public platforms. This is one reason why the occasional negative tweet can go viral so easily: it stands out. On the other hand, voice transcripts from customer service calls tend to be fairly negative. This is unsurprising, because people don’t generally call for customer service when they are ecstatic about your products.

One challenge you are likely to encounter in an analysis of unstructured data is that some of the information doesn’t convey sentiment:

  • The notes that your customer service agents take during service calls may be text-based, but they are meant to only contain factual statements, a report of proceedings. They contain an account of what the customer called about; they don’t give you the customer’s own words.
  • Similarly, maintenance logs can give you a report of what a cleaning crew did on your airplane, but they don’t give you much in the way of emotional content.
  • Free-form survey results might not generate much, depending on the topic and the way you’ve worded the questions
  • Some sources of data, particularly in social media, have a lot of “noise” – text with no real content. Ads, spam, memes, and re-tweets fall under the category of noise.

The key is to use multiple sources of data together – as many as you can, but also to weed out irrelevant comments. Structured data will give you the demographic details. Text analytics will tell you what topics are being discussed. Sentiment analysis gives you the emotional landscape from the sources where it is applicable.  Then you will know when sentiment goes south, what is driving that decrease, and who is unhappy.

Neglecting any of these areas will give you an incomplete view of the customer story. Neglect your customer’s feelings, and they’ll be gone, gone, gone. (Whoa whoa whoa).


Lisa Sigler is Sr. Manager of Content Marketing at Clarabridge. For over 16 years, Lisa has used her writing and editorial skills to bring the value and benefits of technology to life. In her current role, she works to demonstrate Clarabridge’s position as thought leader and trailblazer in the Customer Experience Management market. Lisa holds a B.A. of English from Kent State University. Read more from Lisa on Twitter @siglerLis.
2016-1486_clarabridge_demo_request_cta