Intents analytics gives you data on the intents that are recognized with the highest and lowest confidence by the NLP. These plots can help you better understand where the dataset can be improved. An example: If one intent is often recognized below the confidence level, you might want to look at the variations and user questions for that certain intent to check if it can be improved.


The “highest recognized” shows you the most popular intents that are recognized with a confidence score above the threshold (remember that you can define the confidence threshold for the bot within the dataset settings).



The “Lowest Recognized” shows you the top intents that are below the confidence threshold, thus not recognized as an intent. 

Be aware: Both plots for highest and lowest recognized can index the same intent. This happens if there are a lot of both recognized and not-recognized inputs pointing towards the same intent. This often happens for popular intents, e.g. in this case where the “greeting” message is displayed in both plots. 

Finally: remember that you don’t always want to detect an intent, as it could be a “false positive”.