Explorer: Subclustering

Subclustering is the process of breaking existing groups/clusters of data into smaller ones based on their similarities, trends and existing patterns.

📘

Why subclustering can be beneficial?

Subclustering helps analyse data in a deeper and more fine-grained way. An example would be having a main cluster on water leakage. This can be broken down into sub-clusters such as leakage in balcony, leakage in bathroom, leakage in the kitchen, etc. Such a break-down can make data analysis much easier and more insightful.

Relevance AI's platform provides you with a no-code workflow to subcluster your clustered data with a few clicks. You can access the workflow directly from an Explorer Category view set on your clustering results as well.

How to Subcluster an existing clusters

1. Make sure your AI Clustering is finalized.

   🚧

   Sub-clustering is only available for the One-to-One variant of clustering

   The subclustering option explained below will not appear if your parent clusters are of one-to-many type. Read more about the clustering variants on AI Clustering
   Next, select an AI-generated clustering result on a category/cluster view tab/page. You can label your clusters or skip that step.

2. Scroll down and locate the cluster that you wish to break into smaller clusters (i.e. subclusters). Use the setting on top right, located next to "Ask Relevance", as shown in the image below and hit "Suncluster category".

A window will open asking for the number of subclusters/subcategories. Enter your desired number and hit "Generate subcategories". Wait till you receive a workflow-finalized email notification. You can check the progress in workflow history as well.

Once done, you will be notified on the dashboard with a message too. Click on the link to see the results.

Image below shows how a sample and how five generated subclusters are accessible. Simply click on each button and check the corresponding documents.

Note that you can auto-label the subclusters using "Label your subcategories".

You can also apply subclustering to your whole dataset using the subclustering workflow.