Sentiment analysis extracts subjective information in text. Identifying human emotions in social media is becoming a key means for marketers to better understand their customers.
L’Atelier: What type of sentiments can be analyzed through sentiment analysis?
Tom Reamy: With sentiment analysis, you can identify basic emotions: joy, sadness, fear, anger, surprise, disgust. The challenge is to determine how they are expressed in social media and in text. To do so, you rely mainly on keywords. But it rarely matches entirely what people feel. The key to me is how you go beyond those simple emotions and look at more complex feelings. Indeed, most of the time we are not simply happy or sad, but some combination of the two. A single sentence might have two or three different emotions in it. For instance “I would love your project except for this one feature”: is that positive or negative? We have to go beyond simple emotions to identify more complex sentiments.
L’Atelier: Any examples of insights you can typically get through sentiment analysis?
Tom Reamy: All sorts of insights can be discovered. For instance, sentiment analysis has helped discover fundamentals characteristics of how liberals and conservatives make judgment and how they react to the world. You can analyze that in a text. Indeed, if you analyze how frequently someone uses disgust words, you can come to a fairly good characterization of liberals and conservatives. Conservatives are much more susceptible to the emotion of disgust. And this is confirmed by the analysis of brain scans. You find that when conservatives see something, they are more likely than liberals to have the emotional part of the brain activated.
L’Atelier: How can you turn those complex sentiments into data?
Tom Reamy: With good software you can build rules that take into account things like the relationship between words and between entities. Sometimes, word order changes the meaning of a sentence. You can also have some irony. All those features, based on an underlying language, have to be taken into account when you build the system. The problem is that rules of the human language can be brutal: they change quickly and sometimes they don’t work anymore. What I would like to see is a software that has a learning capability. Therefore it could learn how to detect emotions by itself and ask for human feedbacks.
L’Atelier: On a marketing point of view, how relevant is it to rely on sentiment analysis rather than on focus groups, with a more qualitative approach?
Tom Reamy: The answer is to put together a hybrid system taking into account the strength of each system. Numbers are not going to replace qualitative understanding. You have to do both. It depends on where you are in the process of understanding your customers. For instance, focus groups are very good at the beginning to try to get overall determination of the characteristics you are interested in. It gives you areas to test, then you can confirm this theory using social media and large numbers. You might have another round of focus groups and discussion, and then test it again. That way you build a better model.