Getting started using Communications Mining™

Automation Cloud users

If you are an Automation Cloud user and have AI Units or Platform Units enabled, you can access Communications Mining through the UiPath® IXP service in Automation Cloud. If you do not have any units, but want to start using Communications Mining, contact your account manager.

To access Communications Mining on Automation Cloud, the following conditions must be met:

Legacy users

You do not need to be an Automation Cloud user to access Communications Mining.

Projects can be thought of as restricted workspaces. Each dataset and data source is associated with a specific project, with users requiring permissions in those projects to be able to work with the data within them. Datasets in one project can be made up of data sources from multiple projects. Users will just require permissions in both projects to view and annotate the data.

For more details on data structure, check Understanding the data structure and permissions.

For Automation Cloud users, every tenant has a Default Project that all users within the tenant have access to. Before uploading data, creating datasets, and training models, it is strongly recommended to create a new project with access limited to only those individuals who require access to that data. Once created, it is difficult to move data sources and datasets into different projects.

To create a new project, follow the steps described in Creating a new project (Automation Cloud).

Strict user permissions control access to Communications Mining tenants, projects, data sources, and datasets. You need to allocate permissions to each user. Permissions can provide access to sensitive data and allow users to perform a range of different actions in the platform. As a result, users should only be given the permissions they need to fulfil their roles. For a more detailed explanation of user permissions, check Roles and their underlying permissions.

Data sources are collections of raw, unannotated communications data of a similar type, for example, emails from a shared mailbox or a collection of NPS survey responses.

Creating a source in the GUI sets up an empty source with defined properties, and data can then be uploaded through the API. The setup of this source can also be done through the API.

Once the source is created, data can be uploaded through:

Datasets are comprised of one or more data sources, a maximum of 20, and the model that you train.

If there are multiple sources in a dataset, they should share a similar intended purpose for your analysis or automation.

When you create a new dataset, you can choose to create a copy of a pre-existing dataset. This means that you copy over the same sources, general fields, sentiment selection, labels, and reviewed examples.

Model training involves creating and training a set of labels, that is, a set of intents or concepts, and messages, that is, structured data points, applied to individual communications within the dataset. As we begin to train the model, the machine learning models within the platform will train in real-time and start predicting where else in the dataset these labels and entities may apply.

Training a model requires a model trainer who knows the data inside out. The model trainer imparts their knowledge to the model by training a small set of training data that represents the dataset as a whole, and empowers the model to make predictions on the entire dataset.

Prerequisites before you start training a Communications Mining model include:

The model training process consists of the following key phases: Discover, Explore, and Refine. The Train feature provides a guided training experience that walks users through each phase of training step-by-step.

Any model used in production needs to be effectively maintained to ensure continued high performance. This includes preventing concept drift, and creating an exception process.

For more details on model training, check the following resources:

The platform has a built-in reporting and analytics capability that can help you identify potential issues and improvement opportunities across your communications channels. For example:

For more details on generating insight and building reports, check Using analytics and monitoring overview.

The platform enables downstream automation by creating a queue of communications that a robot can read.

Confidence thresholds levels drive these queues. Setting a threshold means that for the message to enter the queue, the platform must predict that label with a confidence that is equal to or greater than the threshold you set.