About the Communications Mining activities

Core Activities Package. How-To Guide

The Communications Mining core activity package offers cross-platform activities. You can build Communications Mining use cases, such as data uploading from any source, generating predictions for any message, consuming messages from streams.

Glossary

• Project - used to grant permissions to a collection of resources, all resources in Communications Mining are scoped under a project.

• Dataset - the main workspace for all use cases. In a dataset you can train models, save model versions, discover insights, and define streams.

• Source - a collection of raw messages, the underlying data on which models are trained and predictions are generated. Sources can be added to one or more datasets.

• Stream - an API endpoint which serves as a queue of messages with their associated predictions from a model. To define a stream, you must save a model, and the version referenced in the stream.

Cross Platform Activities

Communications Mining Core Activities

The package includes the following activities:

• Create Record - this activity allows you to map any source message to the Communications Mining format. It currently supports two data types: Email and Other. Use Email for email data, and Other for additional data like feedback or reviews. The Create Record activity creates an object for uploading via Upload Record(s) or for making predictions, using Predict Record(s).
• Get Record - allows you to retrieve a raw message from Communications Mining using its ID. Get Record does not offer predictions.
• Upload Record(s) - allows you to upload messages to a Communications Mining source. Start with the Create Record activity to map your data, then use Upload Record(s) to upload it.
  Note: Each message created or updated using this activity costs one AI unit.
• Predict Record(s) - allows you to get predictions from a Communications Mining dataset. Start by using the Create Record activity to map your data. Then use Predict Record(s) to generate predictions.
  Note: This won't upload the data to the platform. Each message prediction request will consume one AI unit.
• Fetch Batch from Stream - allows you to fetch a batch of messages from a stream, with the output displaying model predictions. You must configure Streams in Communications Mining. Here, you can define labels and confidence scores for predictions to be included for downstream processing. You must upload messages to Communications Mining before you can access them via a stream.
• Advance Stream - allows you to update the queue position to the last processed message. Put this activity at the end of your workflow, to make sure that all messages are processed and that you can reprocess any messages in case of downstream failures.
• Reset Stream - allows you to set the starting point for a stream. This can be useful for testing your workflow.