Ibexa DXP Features
Personalization
Delivering Relevant Content
Personalization is all about providing the most relevant content or product recommendations to targeted audiences or individuals in real-time, based on their behavior or demographic data.
Delivering relevant, targeted content and product recommendations in your digital channels is one of the most straightforward ways to improve customer engagement, build brand trust or boost your sales.
The Law of Large Numbers
Ibexa's Personalization Engine collects user tracking information and provides pre-calculated context-based and personalized recommendations.
The background for recommendation computation relies on two aspects, the wisdom of the crowd and the law of large numbers.
A decision that was made by similar users who were previously in the current decision stage is used to provide recommendations. The notion of looking at similar users has emerged from the idea that those who agreed in the past, tend to agree again in the future.
Ibexa Personalization is available for all Ibexa products, you can try it at no extra cost. But, to use it in production, you must purchase the service. Pricing is based on number of personalization API recommendation calls.
How to Configure Good Recommendations
The most important step to get good recommendations is to “understand your data and users.” Each data set is different, and each user behaves differently, however there are certain trends of behavior that you can employ in your recommendation strategy.
To understand the data, it is helpful to start with some basic statistics like the total number of items. An 'item' is an abstract term and, for example, can represent a product, music, an image or a news article for which recommendations should be provided.
If there are different items in the same installation as news articles and news videos these different items should be analyzed separately to understand the different consumption patterns.
Other important statistics to look at are:
- How often content changes
- Frequency of usage: are the items consumed daily, monthly or only once?
- Distribution of usage: Are only a small portion of items used by almost all users (longtail problem), or all items are used equally?
- What are attributes of an item that deliver significant information?
- Natural clusters: Can the data be divided into meaningful groups along with attributes or metadata?
Strategy, Scenario and Rulesets
A scenario is used to define the sequence of strategies that should be used for retrieving recommendations. A scenario can basically accept requests for recommendations from different places on the website, e.g. from a landing page where no context items are available or from a detailed page where a context item can be used for calculating the recommendations.
Rulesets are used to combine a set of Rules that consist of a set of Conditions. The Conditions are a single Filter definition like “no recommendation that have already been shown to the user” or “no items that are cheaper than €10”. To combine single conditions, a Rule is defined and this Rule (or many Rules) are added to a Ruleset which is finally attached to a Scenario.
More details: Scenarios
Models
Models are used to store pre-calculated recommendations. Out of the box, Ibexa Personalization engine offers set of models split into four different types:
- Statistical/ Popularity models – most popular content based on different events, like clicks, purchase or rate
- Collaborative models - calculates similarities in content usage (for example also clicked, purchase together, ultimately bought)
- Editorial models – editors' pickup models (these are created manually), used as a fallback model in the scenario strategy
- Profile models - recommendations based on the previous activity of a user.
All models calculate recommendations for a given time frame (events considered to build models). This can be set up from the UI perspective.
Submodel and Segments Concept
A Submodel is a pre-filtered part of the model. Filtering is done during the calculation using chosen items’ metadata attribute and attribute values. In other words, the model is divided into groups and the recommendation can be provided from the specific group rather than from the main model. Each model can contain several sub-models.
When Sub-models deliver recommendations based on items’ attributes, the Personalization Segment feature does that for the specific users' group. The knowledge of how the content is consumed by specific user groups can improve customer experience, increase visitor engagement and conversion rates.
