What are feedback lists and how are they used?

Overview

Stratifyd allows users to create feedback loops for their custom models. The models include:

• AutoLearn

• Semi-Automatic Taxonomy

• Random Forest

• Feedforward Neural Network

• Logistic Regression

• Embedding attentive model

• Support Vector Machine

• ZSL

These models (excluding Semi-Automatic Taxonomy) are built from the Models Page. Semi-Automatic Taxonomies are built directly from workspaces, where deploying a Taxonomy Model from the Settings tab allows users to select whether they wish to use Traditional Taxonomy labeling or Semi-Automatic labeling. More information on AutoLearn (and other custom models) can be found here. Information on Taxonomy Analysis and Semi-Automatic taxonomies can be found here.

Creating Feedback Lists

Feedback lists can be created using the output of a custom model in the Data Tab of your workspace. The example below creates a Feedback List for a Semi-Automatic Taxonomy.

First, navigate to the Data Tab and find the column for Semi-Automatic Taxonomy labels.

For each row, you will see the labels assigned to each record by the Semi-Automatic Taxonomy. You can add or remove labels to create a Feedback Lists.

The Feedback List can then be used to retrain your model. Doing incremental feedback over time allows your model to improve iteratively, making your model more accurate.

Retraining

A best practice is to schedule retraining for your custom model. This is especially important if you continue to develop your feedback list over time.

Retraining can be scheduled from the Settings Tab of your workspace.

In the Settings Tab, open Advanced Setup for the Semi-Automatic taxonomy you want retrain with a Feedback List. Open the Tuning tab.

Scroll down the page to “Feedbacklist” and select the Feedback List for your model. It is highly recommended that you also schedule model retraining on a regular cadence to use your feedback to iteratively improve the model.

Once you have added your Feedback List and set a retraining schedule, click “Rerun” to incorporate your settings changes.