My professional-life's work has essentially revolved around text analytics, which is awesome! I've had such an amazing opportunity to dive into the industry as CX professionals are realizing the value it holds. Thus, the "happy" from my blog's title.
On the other hand, text analytics is often be wrought with challenges and frustrations, especially in the realm of sentiment analysis, making my professional life a bit more difficult. Hence, the "sad."
My guess is that sentiment doesn't just play with my emotions; you've probably experienced this conundrum, too. I know that several attendees at Seth Grimes' Sentiment Symposium this year expressed similar feelings. In both formal presentations and informal side conversations, I talked with other end-users about how there are so many confusing avenues to working with sentiment, and how looking at it in a looser context is probably the key to feeling happy with the results.
Seth recently wrote an article that reminds us of where sentiment analysis currently stands, and that keeping a broader context in mind will yield more satisfactory, useful results.
For example, in his third point, he says that "[c]apable sentiment analysis will allow you to go beyond positive/negative scoring." This is something that I covered during my presentation at the Symposium. Just focusing on a rough numerical score assigns too much certainty to sentiment - which can be a disappointment. Language is extremely fluid and diverse: a machine can do a great job at understanding the general content of text, but it needs some additional assistance in defining context.
To make up for some of these weaknesses, Walker is working to expand our focus from simple numerical sentiment; we're planning to pilot an emotional overlay to the data to help provide some of that missing context.
To further "drive home" the point of going beyond just numerical sentiment, Seth then mentions in his fifth point that we should "strive to understand both valence and intensity, and also significance, how sentiment translates into actions." What a succinct way of saying we need more context, right? The numbers don't give us enough of the story. We need to know motivation and intensity and truthfulness of the comment. If a customer says, "I wish you had more products," and another customer says, "If you don't fix this issue I'm leaving," it's easy to identify which customer needs immediate attention. Can a tool identify that with just a number, though?
We're looking into ways of giving more context to our sentiment analysis so that we can drop off the "sad" emotion associated with it. Are you doing the same? What troubles or successes have you experienced with sentiment so far? Do you think additional context is needed, or have the numbers given you what you need? Drop me a line and let me know your thoughts!