Trend Detection: How Networked Insights Identifies and Understands Forming Trends

Brands that analyze social media do so for different reasons, from understanding affinities to tracking campaign conversations. As we mentioned in our last posts on taxonomy and how Kairos® works, we’re updating our Kairos product to help brand marketers make smarter decisions earlier and faster than ever. The ability to join a relevant conversation in real time and create awareness for a product or brand regularly leads to marketing success. But often, brands don’t consider the ability to detect the beginnings of a trend. Trend detection, however, is part of the foundation of real-time marketing.

Traditionally, when a topic is “trending” it means that there is an extremely high volume of conversations around that specific topic. But not everything that is “trending” has outstanding volume. Because of this, it’s important to detect relative differences. For brands that don’t receive millions of conversations, a spike from 10 to 500 posts might be noteworthy.

That unexpected occurrence is what our proprietary trend algorithm detects. We monitor more than 25,000 classifiers at a daily and hourly level. Like a heart rate, these classifiers have a resting rate of baseline and expected volumes. Like a heart monitor, our trend detection algorithm determines when a topic peaks at the wrong time or at a higher rate, and sends a signal when something drastically changes.

How? We let the data decide what’s expected and unexpected and perform testing to accurately predict the trend. With enough data points, we can visualize the types of patterns that exist when a trend occurs. When we notice a peak, we can compare it to baseline volume and historical information and test against parameters to determine when a topic is becoming a trend.

For example, the chart below tracks hourly volume for Matthew McConaughey with a baseline measure to understand the expected amount of conversations based on time or day, week and historical information. The peak is clearly above the threshold line, noted in blue. Since conversations around McConaughey spiked beyond the baseline volume and historical pattern, this occurrence is considered a trend.

6-10-2015 11-44-40 AM

The work can’t end once a trend is detected, however. Brand marketers must understand why a trend is occurring. But it’s difficult to wade through thousands of irrelevant posts to find meaningful content.  We make this task easy by automatically identifying unexpected co-occurrences, which are topics in the conversation that do not usually appear together.

Take Starbucks, for example. Not too long ago, there was a significant amount of chatter surrounding the brand. After browsing through our 25,000 classifiers, our trend algorithm detected “racism” as an unexpected co-occurrence – the food industry is not often discussed with topics like race. Digger deeper, the algorithm uncovered that the Starbucks CEO had announced it was okay to talk about race in their cafes.

Using Kairos’ trend detection algorithm, users are armed with more accurate data earlier and faster than ever. To learn more, sign up for our email newsletter at the bottom of this page to receive important updates.