Semantic search for Confluence

[Peter Richter]

I am searching for a possibility to implement semantic search into Confluence. Is there any possibility similar to Mediawiki?

[Aaron Geister · Accepted solution]

Hello @Peter Richter welcome to the community. Confluence is based on keywords. Although you can give labels to pages for easier search and query. 

At the bottom of each page there is the label tab in which you can give the page multiple labels.

This might be the way to help achieve what your tying to do. 

[Peter Richter]

No, that will no achieve the target. E.g when you have a typo like Qality Management you won't be able to find anything. by the way, same issue with Sharepoint...

People are not searching for labels, they are searching for expressions...

A modern search engine will need a semantic search anyway and it's stadard at Mediawiki applications already!

[Aaron Geister]

@Peter Richter I apologize as I didn't understand the semantic type of search. I believe your correct as this is not built into Confluence. 

My only other answer would be to look into the market place for an add-on that might offer this but I am not sure one exist. 

Both Jira and Confluence are both based on keywords and CQL and JQL for query purposes. I am not sure you will find this but if you do please let me know. 

[Peter Richter]

There will be an easy way to implement an incredible better search engine with ElasticSearch (https://www.elastic.co/). Quite easy to achieve, had it with my Mediawiki. no more problem with wrong spelling at search! And the best for Atlassian: I will even pay for that option!

[w_elabbar]

The native confluence search is really bad (keywords). We've gotten much better results by importing into a vector db.

[Peter Richter]

You are right, the search is anno 2010... You just put the keywords into vector db, nothing else? In this case you will need to implement lots of keywords, correct?

[w_elabbar]

It's a bit more involved than that. You will have to vectorize the documents (whole or split into text snippets) and import them along with the document itself into the vectrordb. You will then be able to use semantic search, spellcheck, summarization, etc. 

Your search would then just need to touch upon the concept rather than you having to focus on keyword matches, although you can have a hybrid method where both approaches are used. Check: https://weaviate.io/developers/weaviate.

I'm mainly using it with a genAI chatbot that I developed for Teams:

 

[Peter Richter]

Is this possible also with using at Atlassian cloud platform?

[w_elabbar]

The cloud platform has an API for retrieving documents so yes.