State of the art machine learning to detect competitors sponsored posts

As the organic reach in the Facebook newsfeed is declining, it gets increasingly important to look at advertising strategies to keep your content up in the news feeds. Therefore, when comparing with your competitor it is essential to understand what share of content has been sponsored so that you really have a meaningful comparison. Here,  our new Sponsored Post Detection helps businesses to receive insights in their competitors social media strategy. As the information about whether a post is sponsored or not is part of the private Facebook Insights metrics and therefore cannot be accessed publicly, we are proud to announce a machine learning based approach that can find out (with 96% probability) if a certain post is sponsored or not without the need for admin rights, so for any public Facebook page.

From today onwards, the new sponsored post detection metrics are available in our tool.

In order to get to a really high accuracy, we use a machine learning based approach that gets more intelligent over time. This means, we are not using a static algorithm but we are training the system with actual paid posts so that it can learn how to classify new content. To give you more details on how our vector based machine learning approach works, here are some statements from Sascha and Michael of our developer team:

How does Machine Learning work in general?

Since a while Machine Learning gets more and more attention from companies as it has the power to optimize their processes a lot. Companies are collecting already a lot of data but many lack in using them efficiently for future decisions. Machine learning will help them to get answers to their questions and is kind of artificial intelligence that enables computers to recognize patterns without being specifically programmed. To analyze the patterns you always need to train the algorithm with a sample data set. In our case we used machine learning to recognize patterns of sponsored posts. There are certain characteristics like the number of likes that increases the likeliness of a post to be sponsored. The algorithm can automatically calculate a probability which we then translate into a true or false statement and this is where our Sponsored Post Detection is based on.

What does an accuracy of 96% mean?

Imagine, we use the model to analyze a large set of posts, say 100,000, to get a feeling about which posts are likely to have been sponsored and which are likely to be not. The model will respond by making 100,000 statements like “post number 91,007 is likely to have been sponsored” and “post number 13 is likely to have been not sponsored”. We can be confident that about 96,000 of these statements are right and about 4,000 are false. As we consider larger and larger sets, this confidence will increase, but if we consider smaller sets, it will decrease. So if we would run our analysis on a much smaller set, say 100 posts, we should prepare to see much more false or much more wrong statements than the expected 96.

After testing the new algorithm with many hundred thousand posts and figuring out the highest accuracy we could get, we are delighted to launch our new sponsored post solution today. Knowing now what machine learning exactly is, here you can see some sponsored post detection metrics:

This table shows own posts of the analyzed group as well as a symbol indicating whether a specific post was sponsored or not. So, you cannot only see information on which posts achieved the highest engagement but also if it was potentially sponsored or not.

Beside the table displaying the individual posts we launched a bar chart showing the share of paid and unpaid updates. Hence, you receive detailed information of your competitors social media advertising strategy. Beside that, our newly developed algorithm helps to spend financial resources wisely by promoting the right posts. The bar chart is especially handy when you analyze a longer period of time with a lot of posts.