Analyze conversations using Custom Highlights

Overview

Custom Highlights functions in a similar way as the Smart Highlights feature. A highlight contains key words and/or phrases that CCAI Insights recognizes as being important to determining the user's intent. If highlights are present, they are labeled in the returned transcript. Smart Highlights automatically detects highlights based on a range of pre-set scenarios, while Custom Highlights allows you to specify custom criteria that should be recognized as highlights.

A custom highlight can be made up of one or more phrase match rule groups. A phrase match rule group contains one or more rules. You can use the following parameters to customize the behavior of your rules.

Rules to determine a match

You can specify whether a highlight should be detected if only one of the rule groups resolves to true, or if all of the rule groups resolve to true. Similarly, you can specify that a rule group should resolve to true if only one of its rules is met, or if conditions for all rules in the rule group must be met.

Exact match vs. semantic match

You can specify whether a match must be an exact match to a provided word/phrase or whether a match can be text with a similar intent.

Match type	Description	Additional configurations	Example
Exact match	Determines whether a text is an exact match to a provided keyword or phrase.	case_sensitive. Specifies whether the match must be case sensitive. Values can be true or false.	A highlight must be an exact match to the keyword BERRY. If you disable case sensitivity, BERRY, berry, Berry, and so on will all be considered matches. If you enable case sensitivity, only BERRY will be considered a match.
Semantic match	Determines whether a text has a similar intent to the provided phrase or sentence.	similarity_threshold. Specifies how similar the text's intent must be to be considered a match. Float values can range from 0.0 (extremely unsimilar) to 1.0 (extremely similar).	The phrase "Let me take care of this" must be matched with similar phrases. You can configure the value of similarity_threshold so that the phrase "I will get things sorted for you" is considered a match (with similar intent), but "how are you today" is not.

Create a custom highlight using the Console

1. Navigate to the CCAI Insights Console.
2. If prompted, log in using your Google account.
3. Enter your GCP project ID.
4. Select Conversation Highlights from the navigation menu on the left and go to the Custom Highlights tab.
5. Click Add New, then type in the highlight name and the key words to be detected.

Note: The user interface currently supports a subset of total possible configurations. To access all possible configurations, you must use the REST API.

The configuration in the Console assumes the following:

• There are at most two phrase match rule groups in the highlight and ALL_OF logic is applied at highlight level.
• One of the phrase match rule groups must contain positive conditions and default to ANY_OF logic (represented by Keywords to match). You can add a second phrase match rule group that only contains negative conditions and defaults to ALL_OF logic (represented by Keywords to exclude).
• Keywords are case insensitive.

Analyze a CCAI Insights conversation

After a custom highlight is created, you can use it to analyze your conversations. See the how-to section to learn how to analyze a conversation.

When an analysis has completed, you can view your custom highlights annotations in the Annotations panel along with other types of annotations in each conversation.

In the Custom Highlights tab, you can also find a chart and a table that display the number of each highlight present in the analyzed conversations for different time windows.

Note: A conversation can have multiple custom highlights of the same name. The Custom Highlights feature is applied at the conversation segment level rather than the entire conversation.