5 Steps Defining the Machine Algorithm Flow to Deliver a Customized User Experience

Artificial intelligence (AI) improves as we pave our paths to the future that is Data Driven. Consider Apple’s AI assistant, Siri. And, now think about Amazon’s smart Voice Assistant, Alexa supporting Smart Products. Brands are making us comfortable by making our search easier by featuring what we desire to see. But the use of AI seeps into other industries like eCommerce too. For instance, if you're shopping at Amazon, and browsing through a specific range, the Amazon Recommendation Engine will start prompting you other products that people bought based on user's browsing history, user buying behaviour, location amongst other parameters. 

What could be defined as Machine Learning

Machine learning (ML) is a division of algorithm that enables software applications to act more accurately in predicting consequences without being explicitly programmed. The fundamental proposition of machine learning is to build algorithms that can run statistical analysis to predict an output by receiving input data based on user behaviour. 

Did you know: Machine learning technologies with advanced self-learning capabilities help enterprises execute next-best actions, such as interactive routing, channel orchestration, and dynamic, real-time campaigns/offers.

The General Machine Learning Algorithm Flow

Step 1: Data Collection

While we have the digital data of the organization available, the one existing in the firewall, and the rest of it is achieved from social media or any other third party platform.

Step 2: Data Pre-Processing

When this data is collected at a centralized location. At this stage, a lot of unnecessary data is eliminated as it makes no sense in running meaningful analysis

Step 3: Model Training

This is nothing but a way of further segregation of data and creating various model sets which we apply to evaluate the data further.

Step 4: Model Evaluation

We have tuned the parameters and found the best model that is able to predict the output, and this will now be iterated unless the best model is developed.

Step 5: Improved Performance

The base model returns the number of groups to which particular interaction belongs. These algos are configured to receive the group to which interaction or tags belong. In short, this means the algorithms should work real-time and there shouldn’t be any lag in showcasing the results, which may result in opportunity loss.

Integration of Azure Machine Learning in enabling Personalization

Flavours of machine learning

When the user interacts with the website, it, at the tag collection stage, tags the particular items that the user is visiting are sent in Azure, and based on those tags, the information is clustered in Azure Machine Learning. Now Azure Machine Learning returns back the cluster information as well as items that best relate to those tags. We now get the cluster info. back into Sitecore and based on the information returned by Azure ML, we use the personalization engine that we have built for using rules and other components. This mechanism helps in showing personalised content back to the user.

There are three flavours of ML

1. Supervised: “The outcome against the given input is known before itself” and the machine must be able to map the given input to the output. In this case, multiple images are fed into the machine for training, and the engine must identify the same. Just like a child is shown a picture of a black cat and and when it sees other breeds of cats, it still recognizes it as a cat.
2. Unsupervised: The purpose of this algorithm is to refine the data in some way or to express its structure in a more organized manner by grouping it into clusters.
3. Reinforcement Learning: This is common in robotics, where the set of sensor readings at one point in time is a data point, and the algorithm must choose the robot's next action. Currently, there are no reinforcement learning algorithm modules in Azure Machine Learning Studio.

Looking at Spotify’s customized User Experience

 How Spotify’s Machine Algorithm works in identifying the most suitable track for the consumer based on his music selection behaviour

Under Spotify, users can create their own playlist. Now, picture yourself playing a particular song on Spotify and simultaneously browse through other preferred soundtracks. While this is happening, Spotify’s personalization analysis works in the backend and begins to create a profile for you based on your song listening patterns that could be dependant of – genre, years, artist, mood. This is how Spotify discovers our weekly patterns.

Next Steps

To experience the working of Machine Algorithms and to see how content tagging and user profiling is carried out in Sitecore Experience Database, watch our on demand webinar, 1-1 Personalisation using Azure Machine Learning

Altudo is a Sitecore Platinum partner with 500+ Sitecore projects delivered for 45+ Fortune 500 brands, across 7 industry verticals
We help you realize the true potential of your marketing efforts, speeding up the Time to Value by leveraging CX strategy, 1:1 personalization & our global delivery expertise.