Fallback is a response or behavior that comes into play when a system, model, or bot does not understand a specific question or request. In summary, if the system cannot provide a relevant or even irrelevant answer to the user's question, it is considered a Fallback situation.
For example, if a user writes "About the staff" and is seeking information on this topic, but the chatbot does not understand the question, this is a Fallback (Failure to Understand) situation.
Possible issues arising from this situation include:
- It leads to a poor experience due to failure to meet the user's needs.
- The user may think that the bot is not working correctly and is not fulfilling its function.
For mismatch situations, the fallback intent is activated in our panel, defining the response. This way, user expressions that fall below the confidence levels of the bot are matched with the fallback. If the fallback response is not activated or defined, JetBot will not respond.
If you want to see which intent matches with which expression, you can click on the button with three lines on the JetBot screen and then enter the user expression you want to test.
You can check Fallback situations in conversations through two separate screens in the Jetlink panel.
In the downloaded report, you can view the intents that match user expressions.
You can use the Conversation Number in the report to search for the relevant conversation on the Conversations screen. By clicking the "Train User Expression" link under the Fallback expression in the conversation content, you can open the JetBot Training screen and perform training on the expression as described in the previous step.
If the LLM option is turned off in your bot, meaning it only uses NLP, the "I did not understand" status message described above applies. If the LLM is enabled in the bot, the Fallback response is provided through the LLM. The model first searches for matches using the classic approach. If it does not find a match on the custom-designed LLM Knowledge Base, the LLM Fallback message is triggered.
You can access LLM Fallback messages through the "Responses Generated by LLM" report.
You can use several strategies to prevent fallback situations in chatbots:
Comprehensive Training Data: It is important to train your chatbot with a broad and diverse dataset. Keep your data up-to-date and comprehensive to ensure the chatbot can respond to questions expressed in various ways by users.
User Guidance: Provide guidance and suggestions to help users adhere to certain patterns while interacting with the chatbot. For example, you can use open-ended questions like “How can I assist you?” to guide the conversation.
Integration of Question-Answer (QA) Systems: Integrating a QA system with frequently asked questions and answers can help the chatbot provide quick and accurate responses to such queries.
User Feedback: Collecting and analyzing user feedback can help identify shortcomings and common issues users face with your chatbot. Use this feedback to continuously improve your chatbot.
Ongoing Updates and Improvements: Regularly reviewing and improving your chatbot’s performance helps it adapt to new user needs and language changes.