Not a bug Similar threads widget finding threads only by the same user as thread starter

[Mr Lucky]

Affected version

2.2.2

This can't be right? If so is there way to stop it?

Lesley beach

Hello out there does anybody know the where abouts or what happend to the saxophone player Lesley beach? she played in the band amazula. Stay safe all of ya charlie

cafesaxophone.com

These threads are extremely dissimilar, the only similarity is started by same member (thread start of the thread the widget is displayed in

 

[nocte]

That's not the default behavior (see XF.com) and I am not aware of setting like "only find threads by thread starter".

Look at this thread: https://xenforo.com/community/threads/enhanced-search-2-2-similar-threads.183042/

 

[Mr Lucky]

That's not the default behavior (see XF.com)

Correct which is why I am reporting it as a bug as something is obviously wrong here.

I have tested with all other addons disabled and using default style.

 

[Mr Lucky]

What I have noticed is all the threads start with a similar phrase

Hello does anybody know the where abouts

But those are all stop words. So possibly those are getting included?

I’m also aware of some dissimilar results here:

hScroller issues

When the hScroller is on an element, it appears to be overlaying certain parts of the elements below making it so you can't click them. Adding this to extra.less: .p-body-inner { padding-top: 0; } Will make it so you can't click the breadcrumb links. Or... if you upload an attachment...

xenforo.com

 

[briansol]

I have similar issues... the relevancy of the similar threads is dismal.

 

[Jeremy P]

It looks as though the similar threads on that page are no longer all from the same author. Are you still running into this issue elsewhere? Elasticsearch automatically picks out significant terms from the text based on how (in)frequently they might appear in other messages.

 

[Mr Lucky]

It looks as though the similar threads on that page are no longer all from the same author. Are you still running into this issue elsewhere? Elasticsearch automatically picks out significant terms from the text based on how (in)frequently they might appear in other messages.

Yes, it does look like they now not from same author.

I haven't seen exactly the same issue, but plenty of instances where there does seem to be no relevance still. Also on here as mentioned above.

 

[Jeremy P]

While I can appreciate that the algorithmic selection of keywords can yield results that don't appear relevant in all cases, we wouldn't consider these anomalies a true bug, and the feature does seem to work as anticipated on the whole. We do have some ideas for how we might improve the relevance in future versions.

Feel free to create a new bug report if anything especially odd turns up, but fundamentally the algorithm just picks results that share "more-unique" words in the text body. This can cause it to pick up on the diction of a certain member, which, looking through the results of original post, is the likely culprit here.