Introduction
When users interact with devices, they create hundreds of different data signals, like:
- Search queries
- Websites or apps visited
- Video watch history
These combinations of signals can be used to deliver more relevant marketing to each user using a technology called “machine learning.”
Machine learning
“Machine learning” looks at vast amounts of data to identify patterns and categorize that data into something useful. For example, machine learning can identify valuable users based on the actions they complete. It can then find new users with similar profiles who are likely to complete the same actions. With machine learning, the more data you provide, the better it gets at recognizing patterns you want it to find -- like valuable new users.
Example: Google Quick, Draw!
Let’s look at an example of how machine learning works. Google Quick, Draw! uses machine learning to guess what a user draws based on similar drawings done by previous users. When a user draws a picture of a cat, Quick, Draw! has been trained with a dataset of cat drawings to recognize what a cat is supposed to look like. And with each new drawing, the dataset grows larger, giving Quick Draw! more data to be able to identify new drawings correctly. To understand how machine learning recognizes patterns in the data, let’s see what’s happens to Quick, Draw! behind the scenes.
In order to recognize that a drawing is intended to be a cat, machine learning looks at different data signals. One signal could be the direction in which the lines were drawn. Many people will first draw whiskers on a cat, starting from the face and then moving out and Quick, Draw! can use this to identify if you are meaning to draw a cat.
But machine learning could use additional data signals as well, such as the order of the lines drawn. People might start with the left side of the cat’s face, drawing the top whisker first and then a second or third whisker below that one before doing the same on the right side of the cat’s face.
A scatterplot graph can help visualize the relationship between these two signal dimensions.
Drawings plotted in the blue region are classified as cats and drawings in the red region are classified as something else. Google Quick, Draw! uses machine learning to create a model or set of rules that helps it identify a “boundary” in the scatterplot to separate the cats from the non-cats. For example, drawings with signals to the left of the boundary get categorized as cats, while those to the right are categorized as something else.
What’s really impressive is that Google Quick, Draw! can look at the relationships between pairs of hundreds of different dimensions at once - and set boundaries for each of their scatterplots.
Additional dimensions add more rules to the model that can increase Quick, Draw! ’s ability to accurately predict a drawing.
This is why it’s important for marketers to understand the importance of data when it comes to machine learning. When it comes to marketing data, machine learning works in much the same way as Quick, Draw! It is able to create complex models to make sense of the vast amounts of data about your users and help you more effectively market to each person.
