Feature 1979 conversational experiences #1983
Conversation
rbuck23
force-pushed
the
feature-1979-conversational-experiences
branch
3 times, most recently
from
0552a52 to
42b24d5
Compare
| "type": "object", | ||
| "description": "Rating data for the feedback", | ||
| "properties": { | ||
| "xdm:score": { |
There was a problem hiding this comment.
@cluby-adobe and I have some thoughts around the scoring system we should use for feedback. A couple of options are:
Constrain it to a range of -1.0 to 1.0 as follows:
- 👍 👎 = -1.0 and 1.0
- ⭐ ⭐ = (num-stars / max-stars) * 2.0 - 1.0
- enum (very-negative, negative, neutral, positive, very-positive) = -1, -0.5, 0, 0.5, 1.0
Leave it open-ended and rely on consumer of event to represent/interpret:
- 👍 👎 = 0 and 1
- ⭐ ⭐ = 2
- enum (very-negative, negative, neutral, positive, very-positive) = 1, 2, 3, 4, 5
| } | ||
| } | ||
| }, | ||
| "xdm:feedback": { |
There was a problem hiding this comment.
It's very likely that feedback will come asynchronous to the prompt+response. See the conversation-event examples.
https://github.com/adobe/xdm/pull/1983/files#diff-4ce50d36345c5de493a4373ac7350f5bc09d321bf74b5b68e83a0121f655defcR1
| @@ -0,0 +1,107 @@ | |||
| { | |||
There was a problem hiding this comment.
@rbuck23 , I think it would make sense to just put these fields directly in an agentic-event fieldgroup since they aren't reused.
| "$ref": "https://ns.adobe.com/xdm/common/extensible#/definitions/@context" | ||
| }, | ||
| { | ||
| "$ref": "https://ns.adobe.com/xdm/datatypes/agentic" |
There was a problem hiding this comment.
@rbuck23 , If we want to support the idea of conversation events that are not agentic, it may make sense to have a second field-group that has the agentic information and leave this field-group to just conversation information.
There was a problem hiding this comment.
I like this idea especially as we consider call center transcripts.
There was a problem hiding this comment.
I think what you would want is:
agentic-event.schema.json with properties:
{
"xdm:agentic": {
"$ref": "https://ns.adobe.com/xdm/datatypes/agentic",
"title": "Agentic Properties",
"definition": "Details for agentic event with agents and skills involved."
}
}conversation-event.schema.json with properties:
{
"xdm:conversation": {
"$ref": "https://ns.adobe.com/xdm/datatypes/agentic/conversation",
"title": "Conversation Properties",
"definition": "Conversation event details including prompts and signals."
}
}…tion of conversational signals
… of conversational schema
…sification number -> string
…gle value to an array of string
…rs to be a list of key/value pairs
… signals, add in interaction id, remove utterance-level ids
rbuck23
force-pushed
the
feature-1979-conversational-experiences
branch
from
b700c97 to
f0a4612
Compare
| "description": "Object representing a conversation prompt", | ||
| "allOf": [ | ||
| { "$ref": "#/definitions/count" }, | ||
| { "$ref": "#/definitions/source" }, |
There was a problem hiding this comment.
Could we make sure that prompt has a query property that represents the Raw Text.
| "description": "Object representing a conversation response", | ||
| "allOf": [ | ||
| { "$ref": "#/definitions/count" }, | ||
| { "$ref": "#/definitions/source" }, |
There was a problem hiding this comment.
Could we have something like content property in the response as Raw Text. Also, is the expectation here to always send "text"? There will be use cases when operators will return structured data, we could stringify, but I am just wondering if this is expected.
| "description": "Object representing a conversation feedback", | ||
| "allOf": [ | ||
| { "$ref": "#/definitions/count" }, | ||
| { "$ref": "#/definitions/source" }, |
There was a problem hiding this comment.
Same here could we have a property like content instead of raw or comment
| "title": "Agentic Information", | ||
| "type": "object", | ||
| "meta:extensible": true, | ||
| "description": "Construct to represent agents, agent skills, etc.", |
There was a problem hiding this comment.
@rbuck23 it would be good to add which fields are required, the examples help, but it is not clear which fields are mandatory.
| "title": "Conversation", | ||
| "type": "object", | ||
| "meta:extensible": true, | ||
| "description": "Construct to represent agents, agent skills, etc.", |
There was a problem hiding this comment.
@rbuck23 it would be good to add which fields are required, the examples help, but it is not clear which fields are mandatory.
| "title": "Conversation Signals", | ||
| "type": "object", | ||
| "meta:extensible": true, | ||
| "description": "Conversation entity signals extracted from raw text. Raw text can be from a prompt, response, feedback, etc.", |
There was a problem hiding this comment.
@rbuck23 it would be good to add which fields are required, the examples help, but it is not clear which fields are mandatory.
| "type": "object", | ||
| "allOf": [ | ||
| { "$ref": "#/definitions/name" }, | ||
| { "$ref": "#/definitions/version" }, |
There was a problem hiding this comment.
What does a score for a skill mean?
| "items": { | ||
| "type": "object", | ||
| "allOf": [ | ||
| { "$ref": "#/definitions/name" }, |
There was a problem hiding this comment.
What does a score for an agent mean?
This is a PR to add conversational experiences to XDM. Issue number: #1979