Customer Sentiment: Is Positive and Negative Sufficient?
October 3, 2013
By: Shane Axtell, Lead Computational Linguist, Clarabridge
Your customers are sharing content more than ever throughout their journey. As companies listen to their feedback to extract those relevant insights, they also need to score sentiment in order to accurately capture the customer’s intent.
On the surface, sentiment analysis seems straightforward. In a straightforward world, a simple binary sentiment calculation seems sensible (and that’s how most sentiment analysis systems still do it today). But when you crack the surface and delve deeper into the profoundly experiential nature of people’s lives (which they talk about constantly through social media), the world becomes much more complex and context-dependent.
To confront the challenge of complex human experience, two tools are needed: a scale that reflects the varying degrees of sentiment and methods that accurately present sentiment in context. Some companies have developed tools to accurately capture the varying degrees of sentiment experience. Getting at the context-dependence of sentiment is more of a challenge.
For example, in the following sentence, “thin”, is positive:
My cell phone is thin and sleek and I love it.
However, the same word in the following sentence should be negative:
The walls were thin, making it easy to hear the party going on next door.
Only by utilizing context can a system reliably calculate the sentiment of these words in context.
Your customers are talking about you constantly and they are doing so in the way that comes naturally to them: using complex natural language that gives full expression to their ideas and experiences. If you are able to listen to that complex language, then you will place yourself into a position to meet their needs and respond to their concerns … quickly!