Sentiment Analysis for QA

[CapNcook]

Can someone explain how sentiment analysis works in the context of GPT Connect integration with Caspio? Are there specific settings or configurations required?

[ianGPT]

Hi @CapNcook,

I believe we have the same use-case. I use AI-Integration to automate my QA process and I was also asking AI for the sentiment. Surprisingly the response that I got so far are pretty accurate. No configuration required other than creating your prompt and specifying which tables will be affected.

Here's my sample prompt and response.

Prompt:
 

Here is the chat transcript:  [@field:Chat_Transcript]

Here is the agent name:  [@field:Agent_Name]

Check the holding time of the agent. If the agent placed the customer on hold or didn't reply for more than 3 minutes, provide the transcript where this occurred and the duration of the hold. Do not provide additional information just the transcript excerpt and the exact hold duration.

Check if the chat exceeds 30 minutes and provide the answer as 'Chat exceeded 30 mins: Yes' or 'No'.

Please classify the customer sentiment as Positive, Neutral, or Negative. Respond with: Customer Sentiment: [Positive/Neutral/Negative]

Did the agent demonstrate technical knowledge during the engagement? Answer with 'Yes' or 'No'


Sample response:




Hope this helps!

[Heather]

Thank you for sharing this, @ianGPT!

I have a similar workflow and I am trying to use multiple prompts to populate different fields.
For my Sentiment field, I would like it to be color-coded on my DataPage. Currently, I am just using the following prompt to get the sentiment:
 

Based on [@field:Transcript], provide the customer's sentiment as Positive, Neutral, or Negative.

Is there a better way to do this? any ideas?

[ianGPT]

Hi @Heather,

I just tested it and it was possible. Thanks for the idea!

So what you can do is ask GPT to enclose the answere in a span tag with styling. Ex.

Based on [@field:Transcript], provide the customer's sentiment as Positive, Neutral, or Negative. Wrap the answer in HTML <span> tag with style=”color: green” if Positive, "color:"red" if Negative, and "color:grey" if Neutral.

Then you can render the result as HTML





Here's a sample result of my test.




Hope this helps!

[Heather]

Wow! Didn't know that we can ask GPT to wrap the code in HTML. This is very helpful!