How to improve the Common Intent accuracy?
If you have the User Intent feature enabled on your Workspace you can define Topics and Subtopics based on User Intent.
This feature is powered by our proprietary Machine Learning model, and like most classification models the accuracy of the model will depend on the training data it receives and how it relates to the messages in your workspace.
You can improve the accuracy of the User Intent model with our User Intent Training tool.
How do I access the User Intent Training tool?
You can access it from two places:
- You will see a menu item called User Intent Training on your Left Sidebar. This will take you to an overview page where you can see all available User Intent types. From here you can see your training progress and chose the User Intent you wish to improve the accuracy of.
- Each topic that has a User Intent criteria will display the Improve Accuracy button. You will be taken directly to the training page of the User Intent that the topic is defined by.
If you don't see these options, it's most likely because you don't have User Intent enabled on your Workspace. Please contact us to request access.
How to use it?
With the User Intent Training tool, you can review a sample of sentences and the associated User Intent predicted by Prodsight.
For each sentence, you will be presented with a number of choices:
- If a prediction is accurate, you can confirm that by clicking Yes, correct so the model knows that it should continue classifying similar messages as that particular User Intent.
- If a prediction is inaccurate, you can choose a more appropriate User Intent from the list of available intents.
- If none of the available intents match the sentence, you can specify that by choosing the None of the above option.
Once you make your choice, you will be shown another sentence.
What impact will my corrections have?
Reviewing and correcting examples might sound tedious. However, there are two distinct benefits of using the User Intent Training tool so it's certainly worth investing a bit of time and effort.
- Any corrections you make will have an instant and direct impact on the sentence in question. For example, if a sentence was classed as "Product Issues", but you corrected it to "Product Requests", that particular sentence will no longer appear in topics under "Product Issues" Intent but instead will appear under "Product Requests".
- We use your corrections as the training data to improve the User Intent model on a regular basis. Your corrections will improve the accuracy of the future User Intent predictions.