Contact Us

The wave of personalization and Machine Learning (ML) is shooting up in all domains at a global level. Companies are investing exponentially on these to increase their revenue. Multinational companies such as Amazon, Facebook, Spotify, Netflix, etc. are utilizing personalization and ML to increase customer engagement on their websites. 

98% of customer have disengaged from brands when they don't get a personalized experience

58% of customer have said that personalization is an important factor in determining purchase

11 (3)

"If we have 4.5 million customers, we shouldn't have one store.
We should have 4.5 million stores"

- Jeff Bezos

So what exactly is Personalization? And how to achieve it?
Personalization is the method by which we show relevant content to customers based on needs, demands and their individual preferences, rather than showing the same content to each website visitor.
Personalization is the practice of taking care of every customers needs, demands, tastes and preferences to ensure that their expectations are met and they feel belonged to the brand.

In Sitecore, we perform personalization by setting up rules and pattern cards to show, hide and adjust content on the basis of location, gender, or previous pages visited on the site.

To accomplish personalization using Sitecore, we have to set up rules manually and need a dedicated resource to perform such tasks. A lot of money and time is spent on the resources and software setup for this. Even after spending such a large amount of money, it is very easy to miss the mark with these manual methods. 

Some of the shortcomings of rule-based personalization are given below:

• It is very complex to develop and write rules and set persona for every individual that visits the site.

• In today's world, requirements and needs of users change every now and then, and manual personalization won't perform dynamically

• Even the most experienced and expert individual sometimes miss the minor details of something and can only identify the high-level segments

•New patterns, groupings, segments, and subconscious behaviors will remain undiscovered with rule-based personalization

• It is very expensive to hire experts to incur such needs

But by bringing ML into personalization, we can address most of these challenges. In this blog, we will discuss about machine learning? Flow, of any ML algorithm?, Use of ML for personalization?

 

What is Machine Learning?

Machine Learning provides us the methods and algorithms that can predict and return the output with the help of some input or a set of values as input without being explicitly programmed.
Like we respond to any situation based on our needs and past experience about anything, similarly ,ML helps to predict the response and relevance of something based on the past experiences extracted from the data fed to the ML algorithms. ML is the most scalable means to perform personalization and helps to represent content dynamically to customers. 

 

Why isn’t everyone doing ML-based Personalization?

People do not know where to start and what to do in order to achieve the outcome in the best possible manner. To start ML-based personalization, organizations must identify what to do and why to do. Once this is clear, imbibing ML-based personalization comes handy and so does its maintenance.

Websites generate an enormous amount of data on a daily basis and it is not possible for beginners or intermediates to utilize use of such an exponentially increasing data to perform dynamically.

 

Following are the basic steps for any machine learning algorithm: 1 (2)

Our Approach:

In this demo, we have used Azure Machine Learning studio for performing ML tasks and the Sitecore Cortex for content tagging part. 

2 (2) 

The following figure broadly describes different machine learning and artificial intelligence components. Artificial intelligence consists of machine learning, reasoning, NLP and planning. Machine learning is further divided into supervised, unsupervised, reinforcement and deep learning. In the demo, we are using NLP, Supervised, unsupervised machine learning algorithms.

Need Help?

3 (2)

Following are modules that are developed in this demo:

• Sitecore Cortex (Content Tagging) 

• Data Enrichment

• Data segmentation

• Training and Prediction

• Custom Personalization Rules 

Sitecore Cortex (Content Tagging): From version 9.1, Sitecore provides the capability to tag items by clicking on the 'Tag Item' button available in the ribbon. 

12 (1)

The Sitecore cortex content tagging makes use of third party API-Open Calais. But, somehow open Calais was not working as our requirements. So, we overrode Retrieve Content Pipeline and Get Tags Pipeline and developed our own NLP model using Azure ML studio. Following is the flow of step to develop NLP algorithm:

4 (1)

Clustering and Prediction:

5 (1) 

 

Data gathering: We have used xDB interaction data and the tagging data associated with each and every item obtained from the previous step.

Data enrichment: We have the raw xDB interaction data and Sitecore item tags data. In order to get the best results by applying the machine learning algorithms, there is the a need to convert this raw noisy and untidy data to some enriched form that can be fed to models because output is directly proportional to the quality and quantity of the data.
We have pre-processed raw data, treated missing values, removed highly correlated items, transform and categorize and used One-Hot Encoding techniques for data enrichment.

Data Segmentation: In this step, we have segmented the pre-processed data by extracting the relevant features using 'feature selection' and then applying sweep clustering and Kmeans clustering.

Training: After obtaining the clustered data from the previous step, we have divided the data into two samples-training and testing, and performed model building and model tuning techniques iteratively to get the best model.

Prediction and Recommendation: The model built in the previous step is able to predict the most accurate segment/cluster to which any interaction or behavior of visitor belongs. Then, in order to perform the recommendation task, we have integrated another algorithm which was developed based on the principles of market-basket analysis.

6 (1)

Summary of steps:

7 (1)

Architecture diagram:

 

8 (1)Why should we use AzureML studio for machine learning tasks? 

• Build for cloud-first mobile-first world 

• Reduced complexity 

• Collaborate work with anyone 

• Visual composition, easy to use, drag and drop

 • 500+ Inbuilt R Packages / Python Anaconda toolkit

 • Standard ML algorithms and tools (Text mining, Regression, Classification, Clustering, Anomaly detection)

 • Data sources (Azure Blob, RDBMS, Hive, Web) 

• Data preparation (preprocessing, aggregation, transformation, feature engineering) 

• Deploy in minutes


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.