In this special guest feature, Biljana Belamaric Wilsey of SAS, in recognition of the upcoming holidays, examines sentiments of thankfulness on social media through contextual analysis. Biljana is a linguistic specialist at global business analytics supplier SAS, working to create multi-lingual text analytics applications. In 2015, United Macedonian Diaspora ranked her as one of “40 under 40” rising leaders. Biljana holds a PhD. from North Carolina State University.
At a time of the year that is known for its rampant commercialism, it’s nice to find out that people are thinking about something other than bagging a big-screen TV or gaming console.
How do we know that? By using contextual analysis on the 4,000 tweets posted on a Twitter day called Thankful Tuesday (the Tuesday before the US holiday of Thanksgiving). We found that people didn’t express thanks for an upcoming Black Friday special, massive holiday feast or TV time with a favorite football team.
Instead, the tweets featured words like Thanksgiving, day, family, blessed and love (we were a little shocked that “my dog” didn’t make the top 10).
In about 10 percent of the tweets, people mentioned Thanksgiving, holiday, week, feast and God alongside the hashtag #thankful. The second most prominent topic cluster, present in about 9 percent of the tweets, was time (week) with family, friends and food. All of the topic clusters contained predominately positive sentiments; more than 90 percent were positive. From all of this data, we can generalize that Twitter users generally had happy feelings about these warm topics.
Now one could say that there is a bit of bias in studying a hashtag that seems designed to encourage warm and fuzzy posts. But we argue that without contextual analysis, you can miss the meaning behind what people are saying about your organization on social media. Or even the meaning of complaints made to a call center. That’s a good business practice.
Take the example of a hotel chain worried about negative reviews on blogs and social media. By using contextual analysis, the chain was able to figure out which bloggers/reviewers had the most influence – meaning their posts were shared and linked to. Instead of trying to answer every cranky post, the chain focused on posts from these key influencers. Another example is a theme park that studied tweets to understand the things guests were most confused about and automated the answers to those questions.
A major electronics manufacturer cancelled the redesign of a component after contextual analytics found an obscure, but influential, blog with thousands of comments praising the design of the current component. The company was also able to quickly discover a faulty part by piecing together the two words that emerged over and over again when people called the help desk. The words that contextual analytics picked out were “docking station.” No one called to report a problem with the docking station, the words just came up in those rambling descriptions we often resort to when describing a problem.
Contextual analytics matters because we live in a world where unstructured data (words) accounts for 90 percent of the data available to analyze. Efforts to turn unstructured data into structured data have serious drawbacks. Think about what happens when you need something repaired. The repair person typically tries to select a code for the failure from a standardized list. This information is used to help design better products and proactively seek out problems; but only if the right code is selected.
An appliance company used contextual analysis to study repair notes, call center notes and other structured data to discover the root cause of problems faster. The upshot? The company reduced the service incident rate by 50 percent for products under warranty.
Now that is something to be thankful for.