Produce Automated Response to Service Now ticket based on historical data

I have almost two years of ServiceNow ticket data with all attributes available. I want to create a workflow that will accept a new ticket and respond to it in a customised way based on the training it receives from the historical data. Is there a way to fine tune the historical data and customise the responses in KNIME?

@ipazin , @AFumagali , @armingrudd Can you guys help?

@AnotherFraudUser please see if you can help

Hi @devdataanalyst

Welcome to KNIME Forum. Yes there is a way (as always in KNIME :slight_smile: ) But I assume this is not the answer you are looking for. For me, and I guess most forum visitors, the information you provided is not sufficient to answer this question . How does your input data looks like (sample) what is the output you expect. What have you already tried yourself? See for more guidance.

gr. Hans

1 Like

Hi @devdataanalyst
Welcome to the KNIME Community Forum.

I understand that ServiceNow records incidents, problems, changes,tasks, knowledge articles, etc. So there is a plethora of solutions to your request. You musk cut the elephant in slashes :wink:

I recomend having a look at this article.


@devdataanalyst I do not think there will be a quick solution. One think you could try is:

  • read the article below about large language models
  • convert your existing ticket data into a vector store. YOu might want to indicate segments of the ticket with ‘headlines’
  • find a suitable LLM model from the GPT4All repository (one that is OK for commercial use maybe)
  • edit and test prompts that you could wrap around new ticket texts (give instructions to the model)
  • see if the results doe make any sense

You could also think about doing this with OpenAI if your data privacy requirements would allow this (blog 1, blog 2 ).

You might have to put some effort into such a setup so that it can bring good results.

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.