Predicting posts that are likely to go viral
How would you like to know if a post on Viva Engage or Workplace from Meta is likely to go viral as soon as it’s posted? We've found a formula, based off thousands of interactions, that can help you!
We’re now working on a new feature that will highlight the posts we predict will go viral on the SWOOP Analytics dashboard.
We want to help community managers more proactively moderate communities. SWOOP offers many features that will help with this, for example, we identify top conversations, hot topics, response rates and, at the opposite end of the scale, we highlight posts with no replies.
But there is at least one scenario we haven’t before covered: Is it possible to predict which posts are likely to become popular?
These popular - or viral - posts could be about celebrating a milestone, presenting an innovative new idea, or asking a question that many people care about. It could also be a post about an issue that ignites discussion.
A big challenge faced by community managers is that they can’t predict which posts are likely to “take off”. Instead, they do their best to react once they become aware of a viral post. Bring pro-active means that community managers can better facilitate the emerging discussion, bring other people into the conversation and so on.
Method for predicting high engagement posts
To get a reasonable sample of data to work with, we tapped into our large benchmarking data set. We decided a set of long discussion threads was the best foundation to identify posts that had ended up being “big” (as defined by the number people contributing by reacting or replying in the thread) and thus reaching a large audience. Then we could explore if there was anything special about these posts.
We selected 454 discussion threads which totalled 21,829 messages. Due to our privacy settings, we don’t know who the organisations are, what the conversations are about, or in what communities they are posted.
We came armed with the following hypotheses:
A post by a highly connected person will lead to a discussion thread that has more people involved through replying and reacting than a post by a less connected person.
A post in a large group will lead to a discussion thread that has more people involved through replying and reacting than a post in a small group.
A post with mostly positive sentiment will lead to the whole thread having a positive sentiment.
A post with mostly negative sentiment will lead to the whole thread having a negative sentiment.
Terminology used:
Post: The initial message posted.
Reply: A reply to a post, or a reply to a reply.
Thread: A post with replies.
Reaction: A like, or similar, which can be given to a post or a reply.
We did a correlation analysis on the following metrics associated with each of the 454 posts:
Replies: The number of replies in the thread. Can be a reply to the initial post or replies to other replies.
Reactions: The number of reactions in the thread. Can be a reaction to the initial post, or to replies, or replies to replies.
People in thread: The total number of unique people who replied and/or reacted to the post, plus the initial poster. If a person has replied or reacted several times, then it counts only as one person.
Members in community: This is the count of all the members who were members of the community the post was posted in.
Influence Score: SWOOP Analytics calculates for all people an Influence Score, and it reflects the number of people a person has interacted with over the past three months.
Here is the correlation matrix showing the results:
What we found
Let’s focus only on the correlations that have the highest level of confidence (99%) which are the ones where the p=value is less than 0.001. The conclusion we can draw is:
A person with a high SWOOP Analytics Influencer Score who posts a message in a large community will get a lot of engagement (reactions, and to a lesser extent replies) by lots of different people.
Let’s now turn our attention to the question if the positive, or negative, sentiment of the initial post can be used to predict the sentiment of the replies that follow.
First, we created the scatter plot below to show the placement of each of the 454 posts and associated discussion threads:
You can see there are a lot of threads in the top right-hand quarter which shows most of the neutral/positive threads start with a neutral/positive post. We performed a correlation analysis and when we look at the overall set of 454 discussion threads we can comfortably reach this conclusion: The more positive the post is, the more positive the thread will be.
What about the posts with very negative sentiment? Will a post with negative sentiment also lead to the whole thread having negative sentiment?
If you look at the scatterplot it doesn’t look like it. Some of the posts with highly negative sentiment also led to threads with negative sentiment, but there are also many threads with neutral or positive sentiment.
To be sure about any linkage, we ran another correlation analysis of the posts with a sentiment score of 0.15 or less. Surprisingly, the conclusion is that posts with highly negative sentiment will end up as threads with more positive sentiment, i.e. a negative correlation. However, there were only 38 posts in the sample (8.8% of all posts) so while it is still statistically significant, we can’t be as confident as we were with posts with positive sentiment where we had a much bigger dataset. Still, we can comfortably reach the conclusion that even posts with highly negative sentiment will lead to threads with more positive sentiment.
Conclusion and next steps
We believe we have successfully identified a set of parameters that will be able to predict how popular a post is likely to become. By “popular”, we mean how many people will get involved through replying or reacting to either the post, or to any of the replies, i.e. the whole discussion thread.
Our analysis has shown that with a 99% degree of confidence we can predict:
A person with a high SWOOP Analytics Influencer Score who posts a message in a large community will get a lot of engagement (reactions + replies) by lots of different people.
If the post has positive sentiment, then the thread that follows will also be positive.
Armed with these predictions we believe we can help community managers feel more confident they are aware of emerging discussions. That means community managers can proactively make decisions based on how the discussions unfold. For example, they can tag or inform relevant stakeholders so timely input can be provided, or they can moderate the discussion if required.
Based on the findings above we are now starting to build into the SWOOP Analytics platform a forecasting feature that will highlight posts likely to receive higher levels of engagement across the enterprise, selected communities or topics.
There is another important insight we can draw from this research: Success doesn’t happen overnight. You earn influence through participation. If you aspire for your posts to engage a large group of people, then you need to work for it. It is unlikely a first-time poster will get a lot of people engaged, so you need to build your influence over time.
