Biasing Profiles

Biasing profiles allow you to change the order of results in Search or Browse, by influencing the engine relevancy using the flags that are meaningful to your specific products.

You can bias by:

• an exact value (e.g., bias a specific category)
• number on the record (e.g. bias cheaper items higher, or popular items higher)
• hit against the field during a search (e.g. if hit was against description, lower the relevancy of the product).

Once you create a biasing profile, you associate to an event:

• Area (to be the default for all searches)
• Rule (to trigger in certain conditions)
• via your web application at query time (see the Search API for more details)

Biasing Profile Name

The name you assign to the biasing profile must only contain letters, numbers, or underscores. This name is used in the Biasing Profile selector dropdown in Rules and Area pages.

Influence Slider

The amount of influence or bias the profile will have on search results based on a value from 1-10. The influence slider will define how much the items that you define within the Biasing Profile will override relevancy.

Recommended value

The general recommended value for Influence is 1, if you want to set it at par with innate Relevancy.

Define Bias

You can define biases based off text matches here.

Fill out the Name field with the field against which you wish to bias. You will see autocomplete suggest available fields to you.

You can create two kinds of biases:

1. Exact content value
2. Blank content (search value boost)

Exact content value

Use when: you want to pick out specific products by an attribute: If a field contains a specific value, bias by it.

You cannot add an exact value bias on the title field. If you wish to bias a specific product, please use id, or any other field that’s available for biasing on the record.

For example, if you want to push all the sale products higher, you can use a flag within your records to do a Strong Increase.

Here you see that if variants.onSale is true, we will give it a Strong Increase (products on sale will be higher up in the results).

Note how using the dropdown prevents us from typing a non-existent field.

Blank content (search value boost)

Use when: you want to influence search results. You can make a record more, or less, relevant in the results by adding Increase or Decrease ranks against it.

The way this will work: if a field contained a value that the customer searched for, apply the bias.

Business Use Case

We want to make sure that products that have related keywords are weighted higher to the top, and that products that have a hit against description is weighted lower. This reflects my assessment of these fields as driving my business relevance strategy.

We just need to add both fields with a blank Content field:

Configuration at upload

The field must be searchable for this to work.

Cascade Checkbox

By default, we use all of the biasing items together to build the profile. If you want only one biasing item from the list to apply, set this box to be unchecked. Then, only the first “match” will be triggered. You can reorder the biasing profiles by click-and-dragging them to ensure the order of “triggering”.

Strength of Boost or Bury

Each item in your biasing profile can affect the record by either increasing it’s rank in the results, or decreasing the rank in the results. Absolute increase and decrease values will exert an extreme influence onto the record, where as the other options will work in tandem with relevancy.

Biasing by Numeric Values

Use numeric attributes to bias results. You can do this in 2 ways:

1. Use a field where larger numeric values are better
2. Use a field where smaller numeric values are better

To bias by a numeric field, add it under the Numeric Boost section by typing in the field name. You will see autocomplete suggestions based off all the fields that are defined as numeric within your catalog.

By default, larger values are boosted.

Check “invert” to boost in reverse - smaller values in the field will boost the record higher, and bigger values in the field will lower it in the results.

Records with no value will exert no boost or bury.

The numeric boost is context aware to the query and the results - that is, the values that are used for applying the boost are based off the range within the results of a given query. This means two things:

1. Numeric boost will result in different level of boost, given a specific record-value, depending on what other records are returned.
2. Because this is a context aware functionality there will be a slight performance impact at query time (around 20%).

The exact content of the field that you use for boosting (e.g., whether it contains “1”, or “100”, or “103323432”) doesn’t matter as much as the overall distribution of the values across the results.

Please see the Numeric Biasing article for more examples and detail.

Configuration at Upload

Your field must be defined as a Float at upload time across all Collections.

Default Biasing Profile

It is important to have a general biasing profile for any site as this will allow for the majority of your search pages to be extremely relevant as well as dealing with normal edge cases. Once you create your general biasing profile, you will need to set is as the default for area. Find the selector in Command Center under the Area Management > Edit > Default Biasing Profile.