This July, I'm presenting at The Sentiment Analysis Symposium in NYC. Over two days, industry experts and end users will share insights and learnings about emotion, tone, opinions, and attitudes. More importantly, we'll hear about harnessing the value that sentiment can bring to our businesses and, ultimately, customers.
In my presentation, I'll be discussing the ways that Walker has used sentiment, and some of the benefits and hardships associated with it. Software providers tout its accuracy and ease, while users complain about often confusing or lackluster results. Where does reality fall, though? Is sentiment worth investigating, and can it impact our businesses? I'll give you a sneak peek at how I'll plan to answer those questions during the Symposium:
No user can deny the benefit of near-immediate results and analysis. Within minutes, Natural Language Processing (NLP) can break documents into sentences and keywords, assigning numerical values to each. This is much faster than reading through documents manually and frees up time for working more with customers.
While results are available sometimes with literally the touch of a button, those results are usually trustworthy at just a high level. Automated sentiment assignment can help you get a good feel for the general opinion of your customers – but you're working with "accuracy by volume." The more data you have, the more likely that sentiment will be correct. Once you're looking at handfuls of comments, you might notice small details that the automation didn't correctly assign sentiment to. It got the overall feeling of the comment, but maybe not some specifics.
So … Is Sentiment Worth it?
My opinion? Come to the Symposium to find out! I won't leave you with just that, though; if you make the dive into sentiment, keep these tips in mind:
- You must be patient and work with it aggressively; tools are smart, but not as smart as you. Help the tools learn. Don't blindly trust them; check them over and over.
- You must be willing to extend beyond the tool. Sentiment gives you a nice gist of where opinions stand, but you'll always need to dive into results to find those "nuggets" of insights. Identify an area of negative sentiment, isolate the customers driving it, verify the opinions with other data sources if possible (like anecdotal information, quantitative metrics, etc.), and then dig for your solution.
What are your thoughts on sentiment? Is it your silver bullet, or have you dumped it?