Intent analysis and filtering
The Pega 7 Platform adds intent analysis to its Text Analytics functionality. Through intent detection, you can determine a writer's intent (in posts, comments, messages, and so on) and determine whether the writer is likely to subscribe to your services or buy your products. The intent analysis can also help you prevent churn by reacting to customers who complain or express dissatisfaction about your offerings.
The Pega 7 Platform includes the default pzDefaultIntentModel rule, which detects the following intent types:
Primary | Secondary | Examples |
---|---|---|
direct | cancel |
|
direct | confirmation, yes |
|
direct | confirmation, no |
|
direct | escalation |
|
direct | greeting |
|
direct | stall |
|
inbound | apologize |
|
inbound | compare |
|
inbound | complain |
|
inbound | praise |
|
objective | N/A |
|
outbound | churn |
|
outbound | inquire |
|
sale | purchase |
|
sale | sell |
|
sale | wish |
|
Use cases
Intent detection can be demonstrated in scenarios for chatbot applications:
Scenario 1: Case creation
Case creation based on the detected entities and user intention
In this scenario, based on the text input, the text analyzer that is used by the chatbot can detect all of the information that is required to create a case for booking a flight.
Scenario 2: Incomplete request
Incomplete request that requires further inquiry
If the information that is provided in the input text is insufficient, for example, if the flight destination is not provided, the chatbot might request that the user provide the missing details for booking a flight.
Scenario 3: Escalation to human operator
Escalation to a human operator
In situations in which the user is unable to get assistance, or is annoyed or irritated, the chatbot can escalate and transfer the conversation to a human operator.
For more information, see Intent analysis.
Expression for filtering intent types
You can configure a data flow to filter text analysis content (posts, comments, messages, and so on) based on the detected intent. For example, you can create a data flow that filters out records that the text analyzer associated with the churn intent type. This type of data flow must contain a Filter shape in which the following expression is defined:
.pyOutcome.pyIntents(1).pyName = "outbound > churn"
Previous topic Analyzing the text-based content posted on social media in Pega 7.2.2 Next topic Improving predictions dynamically with adaptive models