Evaluating the Bot Training Model

Evaluating the quality of the training model is an important aspect. It helps you avoid confusing your DRUID bot with phrases irrelevant to the intents or more relevant to other intents.

To access the Evaluation Tool, on the NLU menu, click Evaluation.

To access the evaluation tool in DRUID versions prior to 5.15, select the bot, select solution and from the solution main menu, click Evaluation. The Evaluation page appears by default on the Conversation History tab.

The Evaluation tool allows you to:

• Evaluate the train set per bot language. It uses the train set and evaluate each train phrase. Check if intents are overlapped. This test is not as accurate as using the test set.
• Run test set. Upload test phrases with expected intent and test them in bulk.
• Use actual “real-life” bot conversation as data source to improve the bot NLU model.
• View the words used in your training set and see if there are words used in multiple intents.
• Understand the bot behavior and further fine-tune the NLP model.
• For DRUID version 1.73 and higher, you can also check the matching score of child intents which are shown in the Conversation History, column Flow (the parent flow followed by the matched child intent for that specific message).