Rand Co-Authors Paper on Predicting Optimum Time to Retweet

Most companies and organizations today rely on Twitter to help promote their brand. But with a sea of tweets and other social media posts to compete against, it has become imperative for organizations to know when users are most likely to pay attention to this targeted digital messaging.

William Rand, an assistant professor of marketing with an appointment in UMIACS, recently co-authored a paper studying the timing of user engagement on Twitter.

“Forecasting High Tide: Predicting Times of Elevated Activity in Online Social Media” analyzes patterns of user engagement on Twitter.

The goal, says Rand, is to help predict the best time for social media managers to post content, so that they can maximize the number of retweets they receive.

“For managers who want to take their efforts to the next level, they need to stop relying on simple rules, such as tweet three times of day,” Rand says. “Our model shows that if a marketer really wants to understand when to send out tweets to maximize retweets, they need to pay close attention to both their own individual followers and overall trends on Twitter.”

Rand—working with UMD researchers David Darmon (applied mathematics and scientific computation) and Michelle Girvan (physics) as well as Jimpei Harada (Amazon Japan)—analyzed the content of 15,000 Twitter users.

Users in the study were selected by first choosing a random initial user, then adding their active followers, then the active followers of those followers, and so forth.

In order to model the number of active retweeters at any given time, the researchers used three approaches: a seasonality model that assumes the overall retweet activity on Twitter is primarily dictated by the time-of-day and day-of-week; an autoregressive model that claims retweeting activity is driven by recent activity on Twitter; and an aggregation-of-individuals approach that models the activity patterns of every individual user, and then aggregates these models to describe the overall activity pattern.

The results indicate that both the individual-level model and the autoregressive model did better at predicting the “high tide” of retweeting activity than the seasonality models.

“Depending on whether you have a lot of things to tweet about or just a few things to discuss, the optimal strategy is going to differ,” Rand says. “But no matter what, our results show that you can do better than the simple models that are commonly suggested by social media folk wisdom.”

Rand and his colleagues will present their paper Aug. 26 at the 2015 IEEE/ACM International Conference on Advances in Social Network Analysis and Mining (ASONAM) in Paris, France.