Pega Platform offers two types of topic models that you can use interchangeably, depending on your needs. Learn more about the differences between keyword-based and machine learning topic models and when to use them.
Keyword-based topic detection
Keyword-based topic detection scans a unit of text in search of topic-specific keywords. Based on the detected keywords, topic detection assigns the text to a corresponding topic.
For example, keyword-based topic detection assigns the sentence My uPlusTelco laptop is broken, need help! to thecategory based on the broken and help keywords that uPlusTelco, a telecommunication company, defined earlier.
Use keyword-based topic detection if you have not fully developed your machine learning models yet, or if the models do not produce satisfactory results, for example, because there is not enough training data. You can later switch from keyword-based topic detection to machine learning topic detection.
Machine learning topic detection
In machine learning topic detection, the topic models teach themselves to categorize text by examining previous and incoming text classification. Based on how human operators classify various texts, machine learning discovers patterns for use in topic detection. You can also provide feedback to active machine learning models to increase the accuracy of topic detection in production environments.
For example, uPlusTelco starts storing incoming messages from customers under relevant categories, for example, by assigning the message I want to buy a new phone to thecategory. The company then feeds the collected data to a machine learning topic model. Based on the training data, the topic model learns how to classify incoming customer messages and then starts to analyze incoming text on its own.
Use machine learning topic detection if you have access to previous customer messages and their corresponding categories, or if you can provide the machine learning model with relevant training data.