Content Search API results are unusable and related minimum score question

[jimthornton]

Hi HS community people, this is an updated question on a post from last week.

I'm hoping someone can help confirm that there is no option for improving the results returned. In pulling search results for a hubspot blog using the content search API, the returned posts, in the order they're returned is pretty unusable.

Steps to reproduce the issue

If I search a two word phrase, it's going to return posts matching either word in the phrase by default, sorted by publish date (i think?). 

Possible solution in better use of the minimum score param ?
I went through the docs a few times and tinkered with how to structure the calls to order the results by relevance to query. The minimum score paramter to reduce the number of returned results, which seems to help with relevance a little by limiting results, though the range it uses or what that means doesn't seem to be documented at all.

Minimum Score (&minScore=) Specifies the minimum score threshold to return a given result. This parameter is intentionally set low by default in order to return many results. Increase this for higher precision, but less recall.

I've tried things like 0.05 or 0.01 or 0 and the results returned change, but I can't recognize any patterns as to why or what has changed. It just seems like it limits the results sometimes from a 0 to 0.1 range.

But I know there is a limit parameter so it has to be doing something different than that.

Alternative solutions possibly?
If there is no good solution to order results with content search API, it's starting to look like our options are:
1. return unusable search results
2. migrate the blog to another CMS
3. use a crawler based third party search engine

I'd really appreciate any help or feedback here!

[bradhardinge]

Having the same issue with minScore and not knowing what value to use to try and reduce the number of results returned. Whenever I seach, the best result is at the top but every other blog is also returned below. Whethere related or not!

[Willson]

Hi @jimthornton 

Thanks for the details on this one here! You mention that this is an updated question from the Forum, can you link me to the previous post so we ensure there is no duplication and contain the conversation around this issue to this thread here? 

To jump into the issue, you're correct. Currently we do not have any supported additional query parameters that are documented in our API docs. 

I'd like to take this one offline to look into this further for you and also take on the feedback to ensure it is directed to the appropriate channels. 

As I work on this, i'll be reverting back to this thread to provide updates. Feel free to add any comments as we work on this one. 

[jimthornton]

Hi Matthew, thanks for the reply and appreciate you checking that about the docs and being willing to help assist.

I had just updated this post, so nothing to delete out there. 

Just given we're limited to the parameters we have, do you know if any further clarity can be brought about minscore? What the range is, how it works.

If there was a way to loop through a couple requests with different minscores to get an optimal number (like it always returns at least 3 results), that might get us in a better position.

[Willson]

Hi @jimthornton 

Thanks for your patience here whilst I worked on this internally! I've been able to partner with our Team in relation to the results being returned when using strings as the term parameter. 

As of now, an update has been pushed to accurately detail the results depending on the content and how it matches theterm parameter. This means we should be seeing more accurate results rather than what we were seeing beforehand. 

As an example, i've tested this using my own content and have searched using a full sentence as my term parameter. This has then returned the exact page where this is found to provide a more accurate result. 

Would you be able to give this a test from your side to see how this is working for you?

I hope this helps!