You are about to train a model using Re:infer. You are integral to this training, as the knowledge and experience that you impart to the model during this process will empower it to provide you with highly accurate, real-time insights on your data, on a large scale. 

 

The training process creates a set of labels that are applied to individual communications, known as 'verbatims', within your dataset. These verbatims can be a variety of unstructured data sources, such as emails, chats, customer feedback surveys, etc. 

 

Collectively, all of the labels applied to a single dataset are known as a 'taxonomy'. When complete, this should capture all of the concepts and intents that are relevant to the specific objective you want to achieve.

 

As you begin to apply labels, the machine learning models within the Re:infer platform will train in real-time and start predicting where else these labels can be applied. As you train the model further, the model can more confidently predict labels for other verbatims with similar intents and concepts.

 

The training process can be broken down into 3 core steps – Discover, Explore, and Refine. At first, it is advised to train Re:infer in this order to ensure that you create a consistent taxonomy and to minimise the time-to-value in gaining invaluable insights from your data.



Next: Objectives & building a Taxonomy