nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Using Rovo to Categorize Cases

Albert Catano

Has anyone been able to get Rovo to categorize pages or cases based on categories and sub categories you provide it.

For example:

I create a confluence page with a list of 100+ pages. I have 2 columns where I categorize and subcategorize 10 of them. I want Rovo to do the remaining rows and catgorize them like I have the first 10.

Accepted answers

Jim Knepley - ReleaseTEAM

I've never done that, but that sounds like a multi-shot (sometimes called a few-shot) prompt. I would try to create an agent to do the classification, including the list page with your examples as "knowledge," and have it classify the remaining pages on the list.

It'll probably take some iteration and tweaking, but I can see it working.

For what it's worth, my attempts to classify concepts in a document ran afoul of a guardrail that Atlassian has in place to prevent using Rovo for "critical decisions."

All comments

Jim Knepley - ReleaseTEAM

It'll probably take some iteration and tweaking, but I can see it working.

For what it's worth, my attempts to classify concepts in a document ran afoul of a guardrail that Atlassian has in place to prevent using Rovo for "critical decisions."

Dr Valeri Colon (Connect Centric)

Rovo can assist with pattern-based categorization, but it doesn’t currently auto-complete tables by example. For now, it works best when given clear instructions in a prompt. You can try asking it to apply the same logic used in the first 10 rows, but results may vary depending on consistency and clarity.

KennPangan

Interesting question. I've experimented with using agents to do similar pattern-based categorization. Like others mentioned, Rovo works better when you give it really clear examples and instructions almost like "few-shot" prompting.

We found that if you structure your prompt carefully and provide enough context (like showing exactly how you want it done for the first 10 rows), you can get pretty reliable results, though it still might need manual review.

Curious to hear if anyone else has managed to fully automate this without needing to manually clean up at the end.