Effect of location on NLU model performance in Dialogflow CX

[mangesh1]

Hi all, 
I have 2 queries related to the issue of NLU model's performance:
1. we have a CX agent whose user base is mainly USA, but since the client is from Asia, the location for the agent is set as Asia Northeast1(Tokyo). Is there any relation between location and the NLU model's intent recognition performance?
2. Does the performance of NLU model change overtime due to interaction logging? In the footnote of this setting, it is mentioned that this setting needs to be enabled for "NLU Model Improvement". What does this mean?

[xavidop]

Hi!

In that case what I would suggest is to create your agent in the global region so it works perfectly from all over the world. There is nothing related to the NLU and the region.

For the NLU improvement, check the differences between standard and advanced NLU:
This can be one of:

Standard: Standard NLU technology (default).
Advanced: Advanced NLU technology. This NLU type works better than standard, especially for large agents and flows. Model training will take longer than training with Standard NLU.
Here you have some documentation: https://cloud.google.com/dialogflow/cx/docs/concept/agent#settings-ml

Best,
Xavi

[mangesh1]

Hello @xavidop , Thank you so much for your response. 
We are already implementing Advanced NLU. Thank you for that suggestion.

2nd part my question was, can the NLU model performance change over time without explicit training? Is there any reinforcement learning involved with the interaction logging? There is no explanation regarding this in the documentation. I am referring to this foot note in the settings where it says,
"Note: some features, such as Experiments(opens in a new tab), export to BigQuery and NLU model improvement, require this setting to be enabled, as they rely on traffic logs. "