In this Article 

We will give an introduction to Keyword Explorer, what the key concepts are, how to add recommendations to cohorts, and what settings are needed. 

Introduction to Keyword Explorer

What is Keyword Explorer? 

Keyword Explorer offers recommendations for creating interest-based cohorts by utilizing machine learning. By Analyzing the similarities and connections between keywords, Keyword Explorer suggests relevant terms when you're building cohorts for a specific topic. For example, when creating a 'Fashion' cohort, it highlights relevant keywords to maximize both reach and relevance. 

How to access Keyword Explorer 

Keyword Explorer is available in two sections of the dashboard:

• Custom Cohort Builder 
• Contextual Cohort Builder 

When adding a new condition (predicate), you can select Keyword Explorer, this will launch a modal with the new workflow. Please see the below screenshot which shows where you can see this new option:

What are the key concepts of Keyword Explorer?

Topics

Keyword Explorer gives you cohort-building recommendations for specific topics. Each search, therefore, always starts with a topic, which is entered into the search bar at the top. You need to hit enter to add a topic to the search. 

You can also add multiple topics to a single search, to narrow it down to a specific concept. For example, if you want suggestions for drinks and specifically cocktails, you can use multiple inputs like 'cocktail' and 'drink'. This distinguishes a 'cocktail drink' from something like a 'cocktail dress'.

A topic can consist of a single word E.g. 'Fashion' or multiple words E.g. 'Tour de France'. If the word consists of multiple words, well try to give recommendations for the topic as a while, but if this is not possible we will look for recommendations for each word independently.

Recommendations

There are multiple columns in the recommendations section. Please see below a screenshot, followed by an explanation of each column:

Name: These are the keyword recommendations for your provided topic. They are similar or adjacent keywords. A keyword can be a word or phrase extracted from something like an article title or description, or it could be an exact match from a list of article categories for example. 

Property: A keyword can appear across multiple pageview properties. The property column shows you all the relevant properties for a given keyword recommendation. If there is more than one property you can click on the toggle to see all the properties for the recommended keyword.

Relevance: The Keyword Explorer model is trained on a vast corpus of text and fine tuned on customer data. During this training, it learns the relationships between words by analyzing their meanings based on how often and where they appear together in text. After training, the model represents each word as a point in a multi-dimensional space—think of it like placing each word on a map. Words with similar meanings are positioned close to each other on this map. To determine the relevance score of one keyword to another (or a group of keywords), the model calculates the distance between their points on the map. The shorter the distance, the higher the relevance. In essence, the relevance score indicates how closely related a given keyword is to other recommended keywords. 

Reach: Keyword Explorer provides an incremental reach number for each recommendation. As you select more keywords, you will see the reach of other recommendations declining. As an extreme example, if you had two keywords that have 100% overlap when selecting one of them, the other keyword would have 0 incremental reach. This allows you to only include those keywords in your cohort definition that will have a tangible impact on the total reach. 

Total Reach

Total reach gives you a total user number for your selected keyword recommendations. Total reach accounts for overlap between keywords and therefore gives you a deduplicated reach forecast.

Filters

You can filter your recommendations against the below dimensions:

Properties: Only include specific Pageview properties in your recommendations set. When you don't select any properties, by default all properties are shown. 

Reach: The reach filter allows you to only display recommendations above a certain reach threshold. Once you have started selecting recommendations the reach filter will act on just the incremental reach you would get with a keyword, instead of the overall reach. 

Countries: The country filter allows you to only display recommendations that have been seen in a certain country, based on the 'geo_info' property. 

Domains: The domain filter will allow you to display recommendations that have only been seen in on (or more) of your domains, based on either the 'client.domain' or 'client.url' properties. 

Breakdown reports 

Once you have selected the keywords you want to use you will have the following three breakdown reports appear next to the 'recommendations' tab:

Breakdown by Device: This report gives you a breakdown across device types for your selected keywords. This allows you to understand on which devices you will reach users with your cohort. 

Breakdown by Country: The country breakdown will show you the relative scale across the top countries for your selected keyword recommendations. 

Breakdown by Domain: The domain breakdown will show you the top domains you will reach with your selected keywords. 

How can I add Recommendations to a Cohort? 

Once you have selected all the recommended keywords that you want to include in your cohort definition, click 'Add' to add your selection to the cohort definition. 

Once you have selected the Add keyword button, this will then create a custom cohort and all the behaviors in line with the keywords you have selected, as an example this is what it will look like in the custom cohort builder:

Where are the Keyword Settings defined? 

Settings are defined on the root workspace and will be inherited by all child workspaces. 

What are the key areas of the settings of Keyword Explorer?

The settings can be found within the Keyword Explorer integrations in the settings section of the dashboard, as shown below:

Minimum Reach

You can define a global minimum reach threshold. This ensures Keyword Explorer never surfaces recommendations below that threshold. You still have the reach filter to further refine this for individual searches. 

Property Selection

This setting allows you to define which Pageview properties should be included in the training of the model and ultimately in the recommendations that are surfaced. If there are certain Pageview properties that you don't intend on using or that you think will not be valuable, it's best to deselect them, this will improve the model's performance as well as make the tool faster for your organization. 

Tokenize

Keyword Explorer can extract individual keywords (which could also consist of multiple words) from fields like the article title or description. We recommend enabling "tokenized" for properties that require this kind of extraction but keep it disabled for fields that already contain individual keywords (like a tags field). This setting cannot be enabled for string array properties.