Cloudbot has the ability to generate an internal audit trail for both what people say to a bot and the actions those statements generate on the bot side. It stores this information in Elasticsearch, so that it can be searched, analyzed, and also displayed with a tool like Kibana. While the current package is tested with the Slack adapter, it should work with other adapters, potentially with some tweaks for the exact names and locations of various properties.
There’s a lot that can be learned by looking at bot interactions. For example, what are the bot’s weak points? What is the bot not understanding and responding to appropriately? Who is using the bot? Are certain users or channels heavy users? What value is the bot providing to people? Is it driving traffic to certain services or sites? Are there bugs in the bot? Is the traffic it is driving expected or is it receiving errors from the services it contacts? And this information only scratches the surface of what you can learn.
Using the information gathered by audit scripts, you can then analyze that data and visualize it with a tool like Kibana. Some of the things we’ve found useful to look at have been:
- a table of incoming messages
- incoming adapter traffic by room, robot name, adapter type, and user name
- outgoing HTTP traffic by URL, host, and HTTP status
- when the most recent interaction with each bot took place
- how many deployed bots are out there
For more details, check out the open-source audit repository.
via developerWorks Open http://ibm.co/2cyX8VV
September 19, 2016 at 10:54AM