v237 Configurable NLU models and scoring

“42” - Deep Thought, The Hitchhiker’s Guide to the Galaxy 🤖
SentiOne Automate version 237 was released on 24.04.2024. This release introduced simplified configuration settings for NLU usage.

You can now select different model types and set custom confidence score . Although complex model should provide more accurate results it performance may vary depending on the use case. Please also take into account that complex model uses significantly more RAM (approx. 3.5GB per NLU) comparing to simple (~3MB).

For better handling of emerging and out-of-scope intents you can now set you own level of confidence score. This way you can manipulate the trade-off between precision and recall . It is also beneficial to handle any biases that model may have or when you want to have more confidence to activate certain intent. Change in confidence score are applied instantly and training NLU is not required.

💊 IMPROVEMENTS

• [NLU] Project changelog shows what seed number was used to train model
• [Voice] Configure redirection number in case of Automate, ASR or TTS unavailability

🐛 FIXES

• [Analytics] Navigating from Analytics to Conversation should now properly pass time range.
• [Conversations] Fix for viewing long metadata values in conversation history
• [NLU] Error in admin panel when downloading NLU cross-validation report
• [NLU] Update flow shouldn't overwrite NLU
• [Flow]Disable post processing of bot messages that could break some SSML tags
• [Webchat] Quick replies aren't visible until user scrolls down in Chat tester and Webchat

📚 DOCS

• [Language support] Updated information on supported languages
• [Built-in entities] Comparison for system entities docs
• [NLU] Comparison of NLU models and information of confidence scoring

Please follow changelog for exiting new features! 👋