A crucial part of most software app projects is logging, and social robot applications are no different. In fact, arguably, logging is even more important since the systems are real-time, multi-modal and the fact that it's extremely hard to predict the behavior of users.
The most crucial logging to do for most interactions is usually dialog-logging. This contains what the robot says, what the user says, what intents the robot classified the user utterance as and what states your dialog flow is in, when each of these events happen. One very common use-case of analyzing logs is to enrich the language model of the skill. In practice, this means analyzing the user-utterances that:
- the system did not understand - i.e no Intent classified
- the system understood wrong - i.e it classified the wrong intent
Further on, you probably want to analyze what actions the robot took based on the intent the system classified.
Setting up dialog logging
Furhat comes with a dialog logger that logs the above events to a JSON file. It also logs audio of the user utterances so that you can analyse the You initialize with
dialogLogger.startSession() anywhere in the flow. Optionally, you can add parameters:
name: name for the session, will be used as folder for the log session
cloudToken: token for logging to the cloud (you can get one at your furhat.io profile)
maxLength: The maximum length of the logging (in seconds), after which it will end (default 600 = 10 minutes).
The logging will automatically stop when either:
maxLengthhas been reached (see above)
- You call
- The skill terminates
On your developer machine, the logs end up in:
- Mac and Linux:
There is currently no way for you to view the logs you push to the cloud, but a log-viewing tool is expected shortly.
Since you might, during your Furhat development, build quite complex flows with many states and transitions, we are also supplying a flow logger that logs all relevant info related to why certain flow transitions happened, what triggers were executed etc.