Google Dialogflow: sentiment analysis results not present for google assistant - dialogflow-es

Question

I am working on a dialogflow chatbot, I am on the "essential enterprise" plan and therefore I have access to sentiment analysis which works fine from the simulator panel on the left of dialogflow console, however when I switch to testing my bot on google assistant, dialogflow fulfillment code stops extracting sentiment score from the request sent by dialogflow.
Looking into the request sent in both cases, google assistant and plain dialogflow, I see that "sentiment analysis" results are in fact missing from the request object once you switch to testing on google assistant.
The two request objects are as the following:
Here is the first dialogflow request, not using google assistant (notice queryTextSentiment node near the bottom )
{
"responseId": "b76b18c6-7640-4322-b8e5-2db74cc22656-b55300fa",
"queryResult": {
"queryText": "Very difficult",
"parameters": {},
"allRequiredParamsPresent": true,
"fulfillmentText": "how do you describe the assignments of this course?",
"fulfillmentMessages": [
{
"text": {
"text": [
"how do you describe the assignments of this course?"
]
}
}
],
"outputContexts": [
{
"name": "projects/labeeb-nlddnb/agent/sessions/d66952d9-05c9-bf8a-e083-fa7ccc0bebdd/contexts/assignments_ctx",
"lifespanCount": 5
},
{
"name": "projects/labeeb-nlddnb/agent/sessions/d66952d9-05c9-bf8a-e083-fa7ccc0bebdd/contexts/difficulty_ctx",
"lifespanCount": 4,
"parameters": {
"course_name": "Machine Learning",
"course_name.original": "Machine Learning"
}
}
],
"intent": {
"name": "projects/labeeb-nlddnb/agent/intents/d2d12691-624a-43cd-8b88-b3c7116831dd",
"displayName": "difficulty"
},
"intentDetectionConfidence": 1,
"languageCode": "en",
"sentimentAnalysisResult": {
"queryTextSentiment": {
"score": -0.5,
"magnitude": 0.5
}
}
},
"originalDetectIntentRequest": {
"payload": {}
},
"session": "projects/labeeb-nlddnb/agent/sessions/d66952d9-05c9-bf8a-e083-fa7ccc0bebdd"
}
and here is the request dialogflow sends when I test from google assistant, and dialogflow doesn't set any sentiment analysis results in this case:
{
"responseId": "ce8600dc-4364-48c2-a85b-3acb4cab589e-b55300fa",
"queryResult": {
"queryText": "Very difficult",
"parameters": {},
"allRequiredParamsPresent": true,
"fulfillmentText": "how do you describe the assignments of this course?",
"fulfillmentMessages": [
{
"text": {
"text": [
"how do you describe the assignments of this course?"
]
}
}
],
"outputContexts": [
{
"name": "projects/labeeb-nlddnb/agent/sessions/ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ/contexts/assignments_ctx",
"lifespanCount": 5
},
{
"name": "projects/labeeb-nlddnb/agent/sessions/ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ/contexts/actions_capability_account_linking"
},
{
"name": "projects/labeeb-nlddnb/agent/sessions/ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ/contexts/actions_capability_media_response_audio"
},
{
"name": "projects/labeeb-nlddnb/agent/sessions/ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ/contexts/actions_capability_audio_output"
},
{
"name": "projects/labeeb-nlddnb/agent/sessions/ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ/contexts/actions_capability_web_browser"
},
{
"name": "projects/labeeb-nlddnb/agent/sessions/ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ/contexts/actions_capability_screen_output"
},
{
"name": "projects/labeeb-nlddnb/agent/sessions/ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ/contexts/google_assistant_input_type_keyboard"
},
{
"name": "projects/labeeb-nlddnb/agent/sessions/ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ/contexts/difficulty_ctx",
"lifespanCount": 4,
"parameters": {
"course_name": "Machine Learning",
"course_name.original": "Machine Learning"
}
}
],
"intent": {
"name": "projects/labeeb-nlddnb/agent/intents/d2d12691-624a-43cd-8b88-b3c7116831dd",
"displayName": "difficulty"
},
"intentDetectionConfidence": 1,
"languageCode": "en"
},
"originalDetectIntentRequest": {
"source": "google",
"version": "2",
"payload": {
"user": {
"locale": "en-US",
"lastSeen": "2019-07-24T16:47:07Z",
"userVerificationStatus": "VERIFIED"
},
"conversation": {
"conversationId": "ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ",
"type": "ACTIVE",
"conversationToken": "[\"difficulty_ctx\"]"
},
"inputs": [
{
"intent": "actions.intent.TEXT",
"rawInputs": [
{
"inputType": "KEYBOARD",
"query": "Very difficult"
}
],
"arguments": [
{
"name": "text",
"rawText": "Very difficult",
"textValue": "Very difficult"
}
]
}
],
"surface": {
"capabilities": [
{
"name": "actions.capability.ACCOUNT_LINKING"
},
{
"name": "actions.capability.MEDIA_RESPONSE_AUDIO"
},
{
"name": "actions.capability.AUDIO_OUTPUT"
},
{
"name": "actions.capability.WEB_BROWSER"
},
{
"name": "actions.capability.SCREEN_OUTPUT"
}
]
},
"isInSandbox": true,
"availableSurfaces": [
{
"capabilities": [
{
"name": "actions.capability.AUDIO_OUTPUT"
},
{
"name": "actions.capability.SCREEN_OUTPUT"
},
{
"name": "actions.capability.WEB_BROWSER"
}
]
}
],
"requestType": "SIMULATOR"
}
},
"session": "projects/labeeb-nlddnb/agent/sessions/ABwppHHtMXYQddZeGRTOy0mDfanYvokXr8s72lTD9omqiMy73G3B0JaA0DwvErTRc0HkvhPTmA-CIcAXuQ"
}
Does anyone have an explanation for that?

Answer 1

Sentiment Analysis is not supported for queries originating from Google Assistant.
You can find this statement in the Settings for your Dialogflow Chatbot on Dialogflow.com in the "Advanced Tab" (just below the Toggle to enable Sentiment Analysis).

Answer 2

In fact, sentiment is not logged either, even outside of G Assistant. Here is the log from Operations->Logging in GCP, which is where Dialogflow interactions are logged when you turn on logging. The log there is not the same as the "Raw Interaction Log" that you get from the Dialogflow UI, unfortunately. It's less detailed.
Note that the interesting stuff is all packed in textPayload - in a very hard to parse text string. All that's there is a "Score" variable, which is the confidence score. Thanks goodness that's in there at the least!
Only other way to grab the sentiment score would be to create a custom fulfillment that runs on every intent detect (which is a bummer b/c it'll slow everything down a bit) - and that logs the score separately.
:(
{
"textPayload": "Dialogflow Response : id: \"100aea48-26f8-46b5-ab3a-e48febb0dd1f-e13762d2\"\nlang: \"en\"\nsession_id: \"2e63bd78-79f3-7b02-6efd-5d9f96e1c382\"\ntimestamp: \"2020-06-28T14:08:49.563Z\"\nresult {\n source: \"agent\"\n resolved_query: \"thank you bye\"\n score: 0.68635064\n parameters {\n }\n contexts {\n name: \"__system_counters__\"\n lifespan: 1\n parameters {\n fields {\n key: \"no-input\"\n value {\n number_value: 0.0\n }\n }\n fields {\n key: \"no-match\"\n value {\n number_value: 0.0\n }\n }\n }\n }\n metadata {\n intent_id: \"e5b0d316-d0a0-40c4-b0c3-b0565712aee4\"\n intent_name: \"Closing\"\n webhook_used: \"false\"\n webhook_for_slot_filling_used: \"false\"\n is_fallback_intent: \"false\"\n }\n fulfillment {\n speech: \"closing intent matched\"\n messages {\n lang: \"en\"\n type {\n number_value: 0.0\n }\n speech {\n string_value: \"closing intent matched\"\n }\n }\n }\n}\nstatus {\n code: 200\n error_type: \"success\"\n}\n",
"insertId": "1k1ucxsfhckbar",
"resource": {
"type": "global",
"labels": {
"project_id": "qualitymanagement-eooojp"
}
},
"timestamp": "2020-06-28T14:08:49.574Z",
"severity": "INFO",
"labels": {
"protocol": "V2",
"request_id": "100aea48-26f8-46b5-ab3a-e48febb0dd1f-e13762d2",
"type": "dialogflow_response"
},
"logName": "projects/qualitymanagement-eooojp/logs/dialogflow_agent",
"trace": "2e63bd78-79f3-7b02-6efd-5d9f96e1c382",
"receiveTimestamp": "2020-06-28T14:08:49.589705622Z"
}