Language Model Management
YourCX automatically generates a dedicated categorization model for every open-ended survey question. These models are trained on actual user responses gathered for that specific question.
Whenever a new open-text response is submitted, it is analyzed for both semantic and linguistic patterns. Once a sufficient volume of responses is collected, the platform creates an automatic classification model to organize and tag the responses.
If some responses have been manually categorized, YourCX uses them to train a supervised classification model based on those predefined labels—enhancing the model's accuracy and relevance.
Account administrators can review all automatically generated models, reconfigure them as needed, or build custom models tailored to specific business needs.
Optimizing Model Performance
YourCX language models can be continuously improved through retraining. If you notice quality issues in how a specific category is being classified, you can enhance the model’s training set for that category by:
Importing real user responses that clearly belong to the intended category — use the built-in data importer.
Automatically generating synthetic responses using a language model to enrich the category with varied and realistic training examples.
YourCX includes built-in tools that allow you to generate as many diverse training phrases as needed for each category. This improves the model’s accuracy and reliability — especially for categories with limited data.
For more details on improving your categorization workflow, read the article: “How to Get Started with Automated Categorization in 9 Steps and Save Up to 60 Hours a Month!”
Models for Specific Studies and Custom Categorization Models
For each survey question with manual categorization, YourCX automatically generates a dedicated categorization model. This model learns from manually assigned labels and classifies new responses accordingly.
However, when you need to categorize responses across multiple questions or surveys, it’s more efficient to create custom models. These allow you to define a universal categorization framework that can:
Be applied to any language
Work across multiple questions and surveys
This approach ensures consistent categorization logic across different studies, enabling better comparisons and insights across departments or touchpoints.
Advanced Configuration of Categorization Models
Each categorization model includes several technical parameters that can be customized to fit your specific needs:
Base language model – choose the foundational NLP model for training
Number of epochs – controls how many times the model learns from the training set to improve accuracy
Loss function settings – use weighted or averaged loss to fine-tune how the model prioritizes categories
Minimum category size – define thresholds below which rare categories are excluded from training
Confidence thresholds – set minimum probabilities that must be met for a category to be assigned to a response
Whenever you update these parameters, you’ll need to retrain the model or re-run the categorization process, depending on which settings were changed.
