With FeedHive's new AI-powered tool, you can now predict how well your post will perform.
With FeedHive's new AI-powered tool, you can now predict how well your post will perform.
In this article we will cover:
When we say "perform", we mean how likely your followers are to engage with your post.
So, the result does not indicate how many likes your posts will get.
Neither does it try to predict if the post will go viral.
Specifically, the score is a measure of how likely your post is to obtain a high engagement rate.
ℹ️ Engagement rate is the number of total engagements on a post (likes, comments, shares, link clicks, profile clicks, etc) divided by the total number of views.
So, in short: The higher the score, the higher engagement rate you can expect.
Even though the prediction tool doesn't attempt to predict likes, total views or virality, it's worth noting, that there is a direct correlation between engagement rate, and the likelihood of your post going viral and getting a high number of likes, comments and views.
We analyzed more than 1,500,000 posts from FeedHive and used it to build a powerful machine learning model.
The reason why used engagement rate as the target variable is because:
If we had trained our AI model on likes or views, accounts with a large following would always out-perform accounts with few followers.
That wouldn't be accurate.
So, in an extremely simplified version, we rank all posts in our database based on their engagement rate and train our AI based on that ranking.
One by one, we are telling the AI that "this is what a highly-performing post looks like", and "this is what a poor-performing post looks like".
After doing that many, many times, the AI becomes quite good at understanding the common patterns that make up a high-performing post.
The result consists of two parts:
In this example, the post has mostly similar traits to posts from the training-data that performed poorly.
In this example, the posts that have similar traits are a bit more spread out, and in the lower-middle part of the spectrum.
In this example, we see more similar traits with posts in the higher-middle part of the spectrum.
In this example, the posts have mostly similar traits to posts from the training-data that performed well.