Our Technology

At Spring Social, we use machine learning techniques to find our clients a bigger audience. We know how to predict with high accuracy whether a person will love our client’s brand, or not.

Machine learning identifies patterns via statistics and finding boundaries in sets of data. It can be used to make predictions, and relies on having lots of data in order to be highly accurate. Thankfully, there is a huge amount of data on Instagram, from over 600 million profiles. We realised that not many people are using this data effectively, and that’s how Spring Social began.

Let’s look at an example to understand how machine learning works.

Harry is the Social Media Manager for a new startup that operates a tutoring marketplace in the UK - called TutorMe. Spring Social finds the Instagrammers most likely to follow and engage with TutorMe on Instagram, and identifies the best time to interact with them.

To do this, we gather data from millions of users across over 25 variables covering demographics, personal tastes, social networks, behaviour and activity patterns. Our machine learning model looks at all this data, and based on its training, identifies whether a user is likely to become a customer for TutorMe, or not. We feed the results back into the model, so that it can adapt. This means that over time it becomes better and better at finding followers and engagers for TutorMe.

There are many machine learning methods, utilising various fields of statistics and mathematics. One simple machine learning method is a decision tree, which is easy to visualise.

For every variable, the model creates a boundary. Data goes to one side of the boundary or other. Each fork in the branch classifies a piece of data as possessing one attribute or another, separated by the boundary. Multiple variables and boundaries are organised by the model to form a decision tree. Each additional fork and branch increases the accuracy. You can imagine Instagram users being sent down the tree, being sent to one side the fork or the other, depending on their data.

The final branches are called leaf nodes. The decision tree will classify each user based on which leaf node they arrived at. We end up with two classifications 1. Ideal users, who are highly likely to enjoy and engage with TutorMe 2. Everyone else.

As more users flow through the tree, the more data the model has to rearrange the decision tree to yield better results. This involves changing the importance of the variables, and the boundary points that send data down one branch or another, creating a unique model with its own rules for each client.

As a Spring Social client, our expert staff set up a model based on your customer profile and the type of customers you want to attract. This provides initial guidelines for the machine learning model, which then evolves over time to become more and more accurate. If your customer profile or target characteristics change, we can make adjustments to the model at any time.