Getting Started with Ranking Attributes

This guide will provide you with the necessary information to enable you to use ranking attributes. This is not a technical guide.

Ranking attributes allow you to execute simple or more complex eCommerce strategies. Applying ranking attributes will enable you to order products using a combination of metrics at the same time, rather than pure sort alone. These ranking attributes can then be applied to category and search pages, product recommendations carousels and modification groups.

Ranking attributes as known in Merchandising Studio, may also be referred to as ranking cocktails.

Overview

Key information about this feature:

Ranking attributes are recipes for ranking the items in your catalogue. Ranking attributes are made up of attributes which are the ingredients and their weights which are the quantities.

• Attributes from your existing data model are the main ingredients of the ranking attributes recipes. Verify that each attribute that you want to use in your ranking attributes meets the following requirements:
  • The attribute has numeric values
  • The base type of the attribute is integer or float
  • The attribute is a product-level attribute
  • The attribute is active in the navigation index

Examples of attributes are:

• Product views
• Conversion rate
• Units sold
• Freshness
• Weights represent the importance of the ingredients for the recipe. The weight of each attribute must be a positive number between 0 and 100. The higher the weight number, the higher the importance it has in the ranking attribute. See below for more information on Weights and how they are configured in Fredhopper. 

Normalization 

What is Normalization?

Normalization determines how Fredhopper transforms the values of the attributes so that they are represented on a standardised scale shared across the attributes included in the ranking attributes recipe. This ensures that each attribute in the recipe is evaluated objectively and contributes as prescribed.

Why is Normalization important?

Normalization allows diverse attributes to be used in a ranking attribute by making them comparable. Ranking attributes allow expressing trade-offs between different product attributes by combining their values into a single value.

For example, we have 5 products with information for their ratings and stock levels:

Typically, attributes used in ranking attributes are measured on different scales: in our example product rating has values 1 to 5, while stock level has values 0 to 2000. These scales need to be made comparable (i.e., the attributes need to be normalized) in order for them to be ranked appropriately. If we did not make their scales comparable, product rating would have virtually no influence on the final outcome.

For example, look at products D and E in the table below: a difference of just 0.5 in the stock level offsets the difference between a 5 and 1 star rating!

How does Linear Normalization work?

By default, Fredhopper automatically transforms all attribute values into a range from 0 to 1. To do so, Fredhopper simply assigns the value 0 to the minimum value of an attribute (the minimum in the entire catalog), and the value 1 to its maximum. For all values in between, Fredhopper keeps their relative positions. For example, because a rating of 2.5 is exactly the midpoint of the rating scale, its normalized value will be 0.5. Because this normalization option transforms different scales linearly into a standardised scale (i.e., an attribute value's relative position compared to all other values stays untouched), we call it linear normalization.

The resulting normalized values, together with the resulting ranking attributes values (i.e., the average of the normalized values), are shown in the table below. Note that product E, which previously was assumed superior to product D, now has a very low ranking attributes value, because both rating and stock level are very low, compared to all other products. 

How does Logarithmic Normalization work?

Different from Linear Normalization, Logarithmic Normalization changes the relative position values on their respective scales. Logarithmic Normalization treats differences between values in a different way than Linear Normalization: Linear normalization assumes that all differences are equally important. For example, the difference between stock levels of 0 and 1 are treated equal to the difference between stock levels of 1000 and 1001. Conversely, logarithmic normalization weighs differences between small values higher than differences between large values.

It is therefore important to know whether a product has 0 or 10, or whether it has 1000 or 1010 units of stock.

Pre-Normalized

The pre-normalized options have been introduced to cover cases where attributes are already on the right scale. Basically, it is a "do nothing" option that requires input attributes to be either between 0 and 1, or between 0 and 100.

For example, ranking attributes that were generated during the data transfer and subsequent data transformation process does not need to be normalized again to be used as part of a Merchandising Studio ranking attribute.

Recommendations for Normalization Options:

Pushing

The pushing setting determines where on the normalization scale to place the lowest and the highest attribute values.

Weight

You will need to specify a 'weight' percentage for each of your attributes. The weight of any single attribute cannot exceed 100%. The sum of all attributes can exceed 100%, but it is advised that you keep the total % of all the attributes to 100.

For example:

Where the business strategy specified is to:

Push High number of Purchases: weight (score) out of 60%

Push High Price: weight (score) out of 30%

Push High Stock: weight (score) out of 10%

= Total Possible weight (score) of 100%

The product ordering is therefore a cocktail of each of these weights when combined:

In this example:

Product CH126 has the highest price but, is ordered 8th because it has a low number of purchases and low stock availability compared to other products.

Product CH133 has the lowest price but, is ordered 4th because it has a higher number of purchases and good stock availability compared to other products.

CH132 has almost; the lowest price, the lowest number of purchases and lowest stock, there for across all the attributes it has been ranked 10th.

Product CH130 will appear first as it has the highest weight combination of the attributes in the cocktail.

Scenarios

Automated merchandising strategies

Fredhopper comes with five built in strategies for you to jump start your merchandising, see details below:

These have been developed through years of experience working with top retailers across the world and focus on common areas that are both relevant and important. These Strategies are ready for you to apply and will automatically adjust product assortment overtime. 

Want to take a look? Head to your Merchandising Studio > Ranking > Ranking Attributes.

Best practice

• Always use a unique name when creating a ranking attribute and always prefix with rc_
• If you set the weight to 0, then attribute will not impact the ranking attribute. You can set a weight to 0 when you need to temporarily disable an attribute in a ranking attribute
• Technically there are no limitations of how many attributes you can add to your ranking attribute, however best practice is to use no more than five attributes, as using more than five will dissolve its overall effectiveness
• The weight of any single attribute cannot exceed 100%. The sum of all attributes can exceed 100%, but it is advised that you keep the total % os all the attributes to 100%

Are you ready to create a ranking attribute? Click here to open a step by step guide.

Your feedback is important to us, please take a moment to let us know if you found this article helpful by clicking "Yes" or "No" in the "Was this article helpful?" section below or by adding a comment.