Sitecore Personalization and Machine learning for anonymous users

Question

We have a website developed using sitecore with no authentication functionality. We want to implement new sitecore features in this website. How can we implement sitecore personalization and machine learning with cortex for anonymous user in better way?

Answer 1

If you asking about out-of-the-box cortex-based personalization suggester: it does not matter whether the user is anonymous or known. For machine learning model it uses visit data from all users (whole interaction repository is used as an initial dataset for ML model).

There is no "better way" in ML problems solving. It depends on concrete task and results that you want to achieve. I just can divide work with sitecore and ML in following steps:

1. You should have 100% clear task (understand what you need to do and how should it work)
2. Determine what type of ML problem is your task (classification/regression/clustering/association/etc.)
3. Prepare dataset for your ML model
4. Find train algorithm that is the best for your task and your dataset
5. Integrate it with sitecore (cortex)

Answer 2

What do you want to do?

Before you dive in, the first thing you need to figure out is the business goal that you want to accomplish. You might not need to use Cortex at all, but until you define what you want to do, you are potentially just throwing money away building something nobody needs that is overarchitected.

If you would like to enhance the site, the first thing to do is figure out what is important to the business. What do they need to do? What will help them? Then start thinking about how to get there.

Rules-based Personalization

In order to do personalization for unidentified contacts, you will want to make sure you use rules that are built on behavior instead of demographics. Things like goals triggered during the session, pages visited, etc.

Personalization Suggestions

Personalization Suggestions, which is part of the Cortex brand, works based off the results of Content Testing. So if you would like to leverage that, you can set up some content tests on a component, let it run until you have enough data for the test, and the suggestions will be made based on the different types of users that have interacted with the test.

Cortex Processing Engine

If you would like to do something more customized with your machine learning, perhaps leveraging the Cortex processing engine, then you are likely looking at developing an algorithm and model to analyze interaction behaviour of Contacts in the xDB. Even if they don't log in, you can have unidentified Contacts with interaction histories. You can then employ standard data science techniques on the data in xDB to develop something that works for your business case.