Real-Time Analysis of Markers

In Rasa Pro, Markers are a powerful feature that allows you to track and extract custom conversational events. With the Analytics Data Pipeline, you can now process these markers in real-time, enabling you to gain valuable insights and enhance the performance of your Rasa Assistant. In this guide, we'll explore how to leverage real-time analysis of markers to track solution and abandonment rates in your conversations.

Defining Markers#

Please consult the Markers section of Rasa documentation for details about defining markers.

Enable Real-time Processing#

To enable real-time analysis of markers in your Rasa project, you need to upload the markers to Analytics Data Pipeline with the rasa markers upload command. This command validates the marker YAML against the domain file to make sure all the slots, intents and actions referred in the file also exist in the domain and uploads the marker configuration YAML to the Analytics Data Pipeline (where they're persisted in a database) and exits. To get started, make sure you have the latest version of Rasa installed. Then, open your command line interface and navigate to your Rasa project directory. Run the following command:

By default, this command validates the marker configuration file against your bot's domain.yml file. To specify a different domain file, use the optional -d argument.

This command should be run whenever there is a change in the marker configuration file. The changes might include addition of new markers, changing an existing marker or removing an existing marker.

Configuring the CLI command#

Visit our CLI page for more information about the command line arguments available.

How are Markers processed?#

The markers YAML file describes the pattern of events for marker extraction. Once the YAML files are uploaded, the patterns to be used for marker extraction are stored in the rasa_patterns table. As the Kafka Consumer starts receiving events from the Rasa Assistant, it starts analyzing them for markers. The Pipeline processes all the events from the Kafka Event Broker and identifies points of interest in the conversation that match the marker. The extracted markers are then stored in the rasa_marker table.

The evaluation of Markers in the Pipeline is similar to the rasa evaluate markers command which can be used to process Markers from the conversations in Tracker Store. Read more about it here

Extracted markers are added to the rasa_markers table in the database immediately once they are processed. Each row in this table contains the foreign key identifiers for the pattern, session, sender and last event when marker was extracted along with num_preceding_user_turns which tracks the number of turns preceding the event at which the marker applied. Check out the Data Structure Reference page for the database schema of relevant tables.