Skip to main content
All CollectionsTaxonomy
Tracked Keyword: Best Practices
Tracked Keyword: Best Practices

How to create great Tracked Keywords on Enterpret

Team Enterpret avatar
Written by Team Enterpret
Updated over a week ago

This article helps build an intuition on how Tracked Keyword predictions work, and acts as a guide for creating great Tracked Keywords. To know more about how to create, edit, and manage Tracked Keywords, you can read the article on managing your Taxonomy.

How is a Tracked Keyword defined?

A Tracked Keyword is defined by the following:

Name: Tracked Keyword name is how it'd show up across the platform. Users can query this name to find the Tracked Keyword later.

Description: A simple textual explanation of the Tracked Keyword. This helps the ML models understand the Tracked Keyword and predict it on feedback records accurately.

Included Phrases: A list of words that your users would use to mention the Tracked Keyword. These phrases are used to find feedback relevant to the Tracked Keyword, before deciding whether the feedback should be tagged, based on the Tracked Keyword's name and description.

Excluded Phrases: A list of words that you'd want to to use to filter out feedback that you don't want the Tracked Keyword to be predicted on. If a feedback has these words, it'd be excluded from the pool of feedback evaluated to tag with the Tracked Keyword.

How is a Tracked Keyword predicted on a feedback record?

A Tracked Keyword is predicted on a feedback record based on its definition (name, description, include phrases, and excluded phrases).

Specifically:

  1. Firstly, ingested feedback is filtered based on the included phrases. Any feedback that does not match with any of the include phrases is not considered.

  2. Then, this set of filtered feedback is further filtered based on excluded phrases. Any feedback with any of the excluded phrases is dropped from consideration.

  3. Lastly, this set of feedback with at least one of the included phrases, and none of the excluded phrases,

Best Practices for creating good Tracked Keywords

Good Tracked Keywords capture "what" users are sharing feedback on. These can be features, competitors, or tangible/intangible aspects about your product.

Best Practices: Tracked Keyword Name

We recommend short names for Tracked Keywords, typically one or two words. Tracked Keyword names with more than three words risk being too specific to effectively categorize feedback.

Some good Tracked Keyword names:

  • "Presentation mode"

  • "Free trial"

  • "Login"

  • "Page Access"

Some bad Tracked Keyword names:

  • Names with categories of feedback explicitly called out

    • "Issue with Dashboard"

    • "Template praises"

  • Very generic, and broad names

    • Application

    • App

    • Product

    • <name of the product/company>

Best Practices: Tracked Keyword Phrases

We aim to tag feedback with the right Tracked Keyword whenever it is mentioned in a useful context for the product. Since users may refer to a Tracked Keyword in different ways, we need to account for variations. This is where phrases come in handy. By flagging the different ways you expect a Tracked Keyword to be mentioned, you can ensure that all variants are tagged, improving the recall of the Tracked Keyword.

You can add as many phrases as you want for a Tracked Keyword.

Some good phrases for Tracked Keywords:

  • Keyword: Payment

    • Phrases: Charges, Transaction

  • Keyword: Refund

    • Phases: Reimbursement, return money, return funds

  • Keyword: Integration

    • Phrases: <different integrations your product supports>

Best Practices: Tracked Keyword Description

Keyword descriptions are used for semantically filtering feedback records retrieved with the help of phrases. Good descriptions ensure that the a feedback is tagged with a Tracked Keyword only if there is meaningful feedback about the Tracked Keyword, capturing its nuances.

We recommend writing detailed descriptions that clearly articulate on the Tracked Keyword. A good description should be longer than ten words, and ideally somewhere between 15-25 words.

For example:

Tracked Keyword: CRM Domain Features

  • Bad Description

    • All of the features that make up the CRM domain

  • Good Description:

    • Features include advanced customizable dashboards, comprehensive contact management with seamless integration capabilities, extensive automation for efficient workflow management, and mobile accessibility.

Occasionally, despite being a feature in the context of your product, a Tracked Keyword can mean something general. In such case, we recommend giving enough details to contextualise the meaning of the general term, for your product.

For example:

Tracked Keyword: Timeline

Description: A column which gives a visual representation of how much time is there to complete an item or project and allows to lay out all important project start and end dates

We recommend avoiding feedback categories in descriptions. Adding feedback categories like issues, improvement, praise or help is not required, since these categories would automatically be associated contextually with all Tracked Keywords.


Did this answer your question?