boost control for Vertex AI Conversation with Data Store limitations?

[harlanh]

Hi, I'm developing a Conversation agent using a Data Store. I have a mix of web content and unstructured content. I'm attempting to use Search Configuration to bury (negative boost) a subset of the unstructured content. I can confirm that I can boost or bury a specific url with a condition like:

url: ANY("https://my.specific/url/here")

But I have hundreds of specific pages that I'd like to bury. Two ideas I've tried that don't seem to work:

First, I tried a * in the spec, but it doesn't work, and the documentation suggests that string matches have to be exact:

url: ANY("https://my.specific/*")

Then, I added a field to the metadata for these documents, which I can clearly see is stored in the data store, in addition to the standard "title" and "url" fields. But, trying to bury using that doesn't work:

source: ANY("badsource")

Any ideas? I do want these documents to be indexed for the bot to access, but I want them to be less relevant than other documents, and I don't want them to often be included as a Source Link in the Data Store Response.  

[xavidop]

Hi,

I was checking the documentation and probably can you try the exclusion operator: 

"filter": "url: ANY(\"https://my.specific/\") AND source: NOT(ANY(\"badsource\"))"

 Best,

Xavi

[harlanh]

Sorry, I think I wasn't clear enough...

Neither url: ANY("https://my.specific/*") nor url: ANY("https://my.specific/") (without the star) work to match all documents with that domain in the URL. If there's a way to do that, I can't figure it out.

And I can't get filtering a custom field such as source to do anything at all. 

I'm aware of the logic syntax, but it doesn't address the problems I'm having.

[xavidop]

I am not sure then 😞

Sorry about that