Introduction to Interaction Analyzer Keyword Spotting

Keyword spotting, a subset of speech analytics, is the ability of a monitoring system to recognize predefined words and phrases in interactions.

For example, you are interested in knowing when customers use the word "frustrated" or other words during an interaction with one of your agents. You define the keywords in the monitoring system, such as Interaction Analyzer, and then put it into operation.

Keywords

• Frustrated

• Upset

• Cancel my account

• Angry

• You're not listening

Keyword spotting in Interaction Analyzer is not limited to the words and phrases that customers say. You can also use it to monitor words and phrases for the agent portion of the conversation.

Customer example keywords	Agent example keywords
Frustrated Upset Cancel my account Angry You're not listening	Help me understand How can I help you Sorry to hear that I want to make things right I can't help you

Interaction Analyzer provides keyword spotting for recorded interactions. You cannot use Interaction Analyzer for non-recorded interactions.

Interaction Analyzer keyword reactions

When a keyword is recognized, the system displays an instantaneous, aggregated score of the conversation in the Interaction Supervisor workgroup queue view. A supervisor can react promptly to the score and participate in the interaction. The following diagram displays the default actions that Interaction Analyzer creates when it spots a keyword:

Using Interaction Analyzer, a contact center supervisor can track the real-time conversations in the Customer Interaction Center system. Each tracked conversation shows duration, the involved agent, spotted keywords, the conversation score, and other details.

You can request that the Professional Services do customization of your Customer Interaction Center system to do other events based on keyword spotting. These events range from taskbar pop-ups and screen pops.

Interaction Analyzer aggregated scoring

Aggregated scoring involves the importance that you assign to different keywords and phrases that you define. This score gives you an idea of how a conversation is going. For example, you want to have a conversation score reduced greatly when the customer is getting difficult or problematic. You can assign point values to the keywords and phrases for both agents and customers as displayed in the following diagram:

Interaction Analyzer constantly accumulates the scores for keywords and phrases in a conversation as a point total. It is through this point total that a supervisor can determine which calls require monitoring or intervention.

You can assign any point value that you want to a keyword or phrase, from +/-1 to +/-100. This scoring method defines which calls need the most attention from the contact center supervisor, who monitors the workgroup queue using a few simple metrics. For example, you can heavily score the names of your competitors if they are mentioned in an interaction. Using IC Business Manager, you can view the real-time scores of the agent, the customer, and the sum of those scores as a total overall score for the interaction.

Interaction Analyzer keyword spotability

When you define keywords, Interaction Analyzer notifies you of the spotability of that keyword. Spotability is the predicted ability of Interaction Analyzer to identify a keyword based on different linguistic and phonetic factors that define the keyword. The result is a green, yellow, or red indicator in the administrative interface. The following list describes each of these indicators:

• Green - This color indicates that Interaction Analyzer does not have difficulty in identifying the keyword.

• Yellow - This color indicates that Interaction Analyzer can identify the keyword, but it could also identify other similar sounding words as this keyword.

• Red - This color indicates that Interaction Analyzer has significant difficulty in identifying the keyword or it could mistake other words for this keyword. For example, having a keyword of "hi" could be identified in other words, such as "highway," "higher," "height," and "hype". Additionally, words such as "thigh," "buy," and "guy" could be misinterpreted as "hi." Genesys recommends that you do not attempt to use or further define keywords that have a red spotability indicator.

If a keyword is too short, Interaction Analyzer can interpret that word within other words as a false positive.

For example, identifying "How can I help you" is much easier for Interaction Analyzer to identify as a keyword than "hi." The following table displays how short words are more difficult to identify than longer words and phrases. These short words can be misinterpreted when their phonetic pronunciation exists in larger or similar words:

Keyword	Example false positives
Hi	Highway High prices Hide Hype Hibernate Hyper Height
Mad	Map Mat or Matt Ma'am Man
Caller	Color Collar Column Calmer

As demonstrated by this example, you think that you are specifying a unique word, but consider how Interaction Analyzer can misinterpret this word.

To increase the spotability of a keyword, use phrases that are either in your agent scripts or that commonly appear in agent recordings. The following table suggests some alternatives to the short words that you want to track:

The more syllables that a word or phrase possesses, the better the spotability factor. Consider all possible situations in which a word or a word that sounds similar exists. A higher spotability factor ensures that you are not receiving false positives in your interaction scoring.

Another factor regarding keyword identification is the colloquial variants of words and phrases. For example, "did you know" is often slurred together in everyday speech as "di-jya-no." For more information on how you can phonetically define these types of keywords so that no instance of that keyword or phrase goes unidentified, see "Interaction Analyzer anti-spellings" in Interaction Analyzer Keyword Entry.

Interaction Analyzer confidence threshold

Through the Interaction Analyzer interface, you can alter the confidence threshold of a keyword or phrase.The confidence threshold informs Interaction Analyzer how certain it must be in the identification of a keyword before it reports that keyword as being found.For example, to determine that a caller is threatening legal action, you want a higher threshold so that Interaction Analyzer does not mistake the word for another.This mistake falsely decreases the scoring of the call.The following table shows how some words could be mistaken for keywords:

Keyword	Misinterpretations	Keyword	Misinterpretations
Lawyer	All your Lower Low year Foyer	Attorney	A tourney A journey A gurney A turnkey A churning
Court	Cork Cord Corn Corps Core	I'll sue	Also I'll soon A suit All soup

Interaction Analyzer keyword analysis processing

This topic explains how Interaction Analyzer spots keywords and reports them to Customer Interaction Center for viewing through an Interaction Supervisor workgroup queue view.

When Interaction Analyzer actively monitors a real-time interaction, it does not attempt to recognize each word as this process demands many CPU resources. Instead, it monitors for only the specified phrases in a stream of words for each channel (customer and agent).

Interaction Analyzer processes each side of the conversation regardless of interruptions or when one person talks over another as seen in the following diagram:

In this example, the customer and the agent are interrupting each other. However, each side pauses speaking after being interrupted as they listen to the other side. It is during this pause that Interaction Analyzer reports the identified keywords, if any.

There is a brief delay of a few seconds between the time when a keyword is spoken and identified, and when it is displayed in the workgroup queue view of Interaction Supervisor. The following diagram displays how Interaction Analyzer processes this conversation and when the keyword is displayed in Interaction Supervisor:

All times displayed in this diagram are approximate. Although there is a brief delay in real-time reporting, this delay does not affect how the keyword is identified in the recording. The keywords in recordings are identified at the exact point where they occur.