On August 21st, millions of people across the United States stepped outside to view the progression of a rare, stunning solar eclipse. During that time social media was flooded with pictures and mentions of the eclipse.
The conversation around the eclipse was one of those rare social media conversations that captivated all sort of users and elicited a wide array of responses and reactions. Naturally, we did what any AI-powered consumer data analytics company would do: Built an image classifier that was able to recognize photos of the eclipse on social media and then compared it to a classifier that captured text mentions of the eclipse.
What we found was quite interesting. Here’s the story of what we learned.
The first thing we discovered was that using the eclipse image classifier boosted volume of eclipse-related posts we identified by 8%.
In other words, our eclipse image classifier allowed us to identify an additional 8% of eclipse photos that were included in social posts without an eclipse-related keyword. Without the classifier, these images would not have been surfaced.
An 8% increase is important, but it wasn’t the important thing. We quickly realized that 88% of the posts with an image and the keyword eclipse in the body didn’t even have a picture of the eclipse. In fact a large proportion were just SELFIES!
Put another way, nearly nine out of every ten social posts that mention the eclipse and include an image do not contain an image of the eclipse itself.
So what’s the takeaway?
People often think of image/logo recognition as way to expand their search or fill in the “blind spots” and tell the whole story. And image analysis is certainly effective at doing just that.
However, in many cases, it’s also a great way to reduce noise and get to the highly relevant conversation that you actually care about. Not simply adding more data, but actually helping you eliminate the irrelevant data and key in on the posts that actually matter to you and your business.
Blocking out the noise
So back to the eclipse for a second.
Say we wanted to analyze the eclipse conversation on social, but what we really cared about is user-generated photos of the actual event (and not millions of selfies of people waiting for the eclipse).
Using the eclipse classifier allowed us to reduce 88% of the noise and focus on content that truly mattered to us…the posts that contained images of the eclipse.
Conclusion: The future of image analytics
Ultimately, we learned that by creating the image classifier we were able to accomplish two distinct but related goals:
- Identify an additional 8% of posts that did not include an eclipse keyword but did have an image of the eclipse.
- Eliminate a full 88% of the posts that included an eclipse keyword but did not have an image of the eclipse itself.
Clearly, image analytics can help brands, analysts and researchers find the social posts they actually care about. But the surprising thing we learned is that advanced image analytics can also help us immediately disregard the posts that seem relevant but actually aren’t.
The future of image analytics is bright. Bright enough that you might not want to look directly at it.
Interested in learning more about image analytics? Download our comprehensive image guide below.