FAQ

Common Questions

How do I report an incorrect classification?

From the Dashboard

From within the dashboard, you can create corrections directly from any API result. If you have log retention enabled, navigate to Spam Logs -> Show within the retention period, and click the "Create a spam report" button. This section also allows you to review and manage all your spam reports.

From the API Response

To report false positives, you can use the mark_as_incorrect URL provided in the response. This URL is public and can be accessed from anywhere, making it convenient for various integration flows.

How much custom training data is required for the fine tuned model to be effective?

We've found the fine-tuned model requires a minimum of around 1000 samples of spam and ham content each to be effective. However, the more data you provide, the better the model will perform om your specific use case.

Can I get started without providing custom training data?

You can start using the SpamCheckAI API without providing custom training data. The model is pre-trained on a large dataset and can provide accurate results out of the box. One approach might be to start with our model running in parallel with your existing spam detection system. Over time, you can collect data and create spam reports within the dashboard. Once you've collected enough data, you can fine-tune the model to improve its accuracy specifically for your use case. For most users, the pre-trained model is sufficient to get started and by opting to include your spam reports, you can help improve the model for everyone.

Are all spam reports reviewed by a human?

All spam reports are reviewed by our team to ensure the accuracy of the model. This process helps us improve the model and provide better results for all users. This is one of the primary reasons to fine tune the model for your needs. Our model is based on our opinion of what spam is which may not always align with your needs.

How long does it take to train a custom model?

The fine-tuning process typically takes around 1 hour to complete. However, this can vary depending on the amount of data provided and the complexity of the content. Once fine tuning is complete the model will be selectable in the dashboard.

FAQ