You can achieve it only by custom code implementation. You would need to handle failed classification requests, collect data and then call model re-train. In my new article, I show how to implement chatbot backend in Node.js, you can do all such custom logic there (in my example, I handle chat context there) — https://medium.com/oracledevs/machine-learning-applied-tensorflow-chatbot-ui-with-oracle-jet-custom-component-c7bb2ea2f8bf

TensorFlow Certified Developer | Machine Learning Expert | Oracle Wizard | Founder katanaml.io