Skip to main content
All CollectionsClassifiers
How to Build a Classifier
How to Build a Classifier

Step by step guidance on how to build a Classifier in Authenticx.

Updated over a week ago

How to Build a Basic Classifier

In the Classifier Overview article, we talked through a simple example of an Agent Quality Classifier: did the agent end the call positively?

Let’s continue with this example as we discuss how to build your own Classifiers.

1. Navigate to Classifiers, and click into Classifier Builder

This is where you’ll build your new Classifier.

Screenshot 2024-08-09 at 4.54.27 PM.png

2. Give your Classifier a name, category, group, and description

Screenshot 2024-08-09 at 4.46.59 PM.png

3. Click + New Group to start building

Screenshot 2024-08-09 at 4.47.47 PM.png

4. Build your Ruleset

Input words or phrases to look for. Choose contains, does not contain, or starts with for each word or phrase. Do not copy and paste from Word or other external documents here. Instead, type your words or phrases directly into the text box.

Screenshot 2024-08-09 at 4.50.06 PM.png

Click the AND button to change the logic to OR.

Screenshot 2024-08-09 at 4.51.18 PM.png

Click + Condition if you’d like to only search for a phrase or group of words said within a certain number of words. Please note tat 5 is the maximum number of words you can search within (e.g, ~5 will search within 5 words, ~6 will result in an error message).

💡 This means you are not searching for the exact phrase, but, for example, the words “thanks” and “business” within 5 words. In this example, if the agent says “thanks for doing business with us”, the Classifier would identify it as a positive ending.

Screenshot 2024-08-09 at 5.04.47 PM.png

5. Test your Classifier!

Choose an interaction date range, and then click Test Rule Set at the bottom of the page to see how well your Classifier works.

Screenshot 2024-08-09 at 4.52.48 PM.png

Here, you can check out conversation summaries and snippets of the transcript that met the Classifier conditions you set up (these will be highlighted blue in the results). For the sake of testing, the builder will only show a maximum of 100 results in this area.

From here, you may choose to refine the Classifier further if there were no search results, or if the rules you set up aren’t quite as fine-tuned as you’d like. You can do this by clicking Edit Search Terms.

Screenshot 2024-08-09 at 4.56.21 PM.png

6. Save your Classifier

Click Save Classifier when you’re ready. This will allow you to access and edit it at will.

You’ll then see the classifier builder page, with the basic logic you input automatically formatted in the Advanced builder area.

Screenshot 2024-08-09 at 4.59.35 PM.png

7. Publish!

Once you’ve tested your Classifier and saved it, you can click Publish. This will make the classifier available to all users of your organization, and it will automatically run your classifier against any incoming interactions moving forward. You can see your Classifier at any time (and access it to edit it) in the Your Classifiers list.

Screenshot 2024-08-09 at 5.26.12 PM.png

8. Want to Un-publish?

If you no longer want to automatically run this classifier against incoming interactions, you can always Un-publish. Simply select the Classifier you want, and you’ll see the option to Un-publish appear. Note that the existing data already found by a Classifier when you Un-publish will remain in the platform.

Screenshot 2024-08-09 at 5.26.44 PM.png

Classifier Results

When we talk about results of a Classifier, we say positive or negative. What does this mean?

  • Positive = a result that indicates the Classifier rule does identify a “hit”

  • Negative = a result that indicates the Classifier rule does not identify a “hit”

When you perform a search using Rule-Based Classifiers, the conversations returned are 100% accurate matches based on the criteria you’ve set.

In addition, you'll see highlighted text snippets, which are designed to provide a quick preview and help guide you to the relevant parts of the conversation. These highlights may not always perfectly align with the exact phrases or words you specified in your Rule-Based Classifier criteria, as they are intended to provide context and help you decide which conversations to explore further.


How to Build an Advanced Classifier

If you’d like, you can build an advanced Classifier by clicking Advanced in the Classifier Builder. This will allow you to define very specific rules for your Classifier, as follows. If you’d like to tap out at this point, don’t worry — you can work with an Authenticx team member to build an advanced Classifier.

There are a few pieces of info you need to know to build an amazing Advanced Classifier. Let’s do this!


Best Practices for Building a Classifier

Avoid Stop words

These are words that will be ignored by your search, because they’re way too common in the English language, and searching by them would give us way too many false positives. Don’t use these words in your Classifier:

image.png

Avoid Repetition

There is no need to include repetitive words or phrases when creating a Classifier.

For example, writing "am unhappy"~4 OR "am very unhappy" ~3 is not necessary. The "am unhappy" ~4 will catch any scenario where you'd see "am very unhappy"~3.

Operators

These are the functions that help narrow your parameters. They're the things that allow you to search for phrases. Here are the ones you need to know:

  • Quotation Marks “ ” : Placing words within quotations will search for those words within sequential order

Example: "Peanut Butter Jelly"
Positive: "Sandwich with peanut butter and jelly"
Negative: "Jelly, crackers, maybe some peanut butter"

When you use quotations, the sequence of the terms (not the proximity or inclusion) dictate a positive or negative identification of the Classifier.
  • Tilde ~ :

    • Placing a number after a tilde will search for those words within # many words.

    • The order of words does matter. The AI will search for each word in your quotation marks within ~x of the next word. For example, "Peanut Butter Jelly"~3 looks for "Peanut" within 3 words of "Butter", and "Butter" within 3 words of "Jelly".

    • Please note that the maximum number of words you can search within is 5 (~5 will work, ~6 will result in an error).

Example: "Peanut Butter Jelly"~4
Positive: Peanut Butter and Jelly Sandwich
Negative: Peanut Butter Crackers, and I sometimes have it with Jelly on the side
Negative: Jelly donuts covered with peanut butter

When you have a tilde, the words must appear within that many words of one another for there to be a positive identification of the Classifier, and they must be in the correct sequence.
  • AND: placing this word between two queries will search for both. This can be used to join phrases and/or words.

Example: "Peanut Butter"~2 AND Sandwich
Positive: Peanut Butter and Jelly Sandwich
Negative: Peanut Butter Crackers and sometimes there is Jelly on the side

The negative does not contain the word SANDWICH, which we included clearly in our Classifier.
  • OR: placing this word between two queries will search for either. You may type this word or press the return key to utilize this operator (each new line serves as an OR operator). This can be used to join phrases and/or words.

Example: "Peanut Butter"~2 OR Crackers
Positive: Peanut Butter Crackers, and sometimes Jelly on the side
Positive: Peanut Butter and Jelly Sandwich

Including the OR allows the Classifier to return positive when either of the terms or phrases is identified in the interaction.
  • NOT: using this operator before a word or phrase will exclude any search results where that word is contained. Use with caution.

Example: "Peanut Butter"~2 NOT Sandwich
Positive: Peanut Butter Crackers, and sometimes Jelly is on the side
Negative: Peanut Butter and Jelly Sandwich

When the term after the NOT is present, it would be excluded from your search return.

Did this answer your question?