ES 2.2 Does the new 2.2.0 B1 'Similar Threads' widget use the threads titles?

Kevin

Well-known member
Does the new 2.2.0 B1 'Similar Threads' widget use the threads titles as a basis for being similar? Any settings I can set so that the title gets weighted higher? 🤔

Below is a shot showing a list of recent threads all in the same node. Note the various threads that have word "Apollo" in their title. When viewing the "Apollo 11: Returning from the Moon" thread and the 'Apollo 11 Launches into History' thread, for example, the various 'Apollo' titled threads don't show.

1595541665194.webp

1595541802517.webp
 
Does the new 2.2.0 B1 'Similar Threads' widget use the threads titles as a basis for being similar?
Yes, but it's also more complex than that. The approach taken is described in the Elasticsearch docs, though it's tough to understand: https://www.elastic.co/guide/en/elasticsearch/reference/current/query-dsl-mlt-query.html

Similar thread result tuning is generally something that we knew we'd need to do post-beta release, as it benefits from tests on various "real" data sets. There is a tool that gives you a bit more info, though I think I may need to expand it to allow some easier query modification testing. If you enable debug mode (it's not accessible outside of that), you can go to admin.php?enhanced-search/mlt-analyzer. From there, you can test the system as it works both for thread creation-based searches and finding threads that are similar to an existing thread.

If you want to do a quick and dirty test, you can set the title and forum in thread creation and compare the results based on URL. This will basically demonstrate the difference between a title-only match and a more general search.

Saying that, simply limiting to the title when we have more information may not necessarily be the best option -- but more analysis is definitely required.
 
@Mike, thanks for the info & the test tool link to check out.

Interestingly the results on the test tool aren't matching what's actually being shown in the widget for at least a few threads. I've tried rebuilding the search index and manually running the cron job for the similar threads, same results. Oddly newer threads created since then are giving better results and match the tool.

Is there a method of wiping the cache entirely and start with a fresh rebuild?

Here's the results showing for the thread in the ACP tool....
1595612130840.png

Here's what the widget shows when viewing the thread....
1595612266618.png
 
I think it could work well for many cases if there is an option to make it just work on titles (I had probably better make that a suggestion). I'm sure it can work technically well with the clever stuff it obviously does, but for users in general, I think it will grab them more if they can see somebody else asked the same question. And on a forum, the question is usually in the title.

And we can see from the similar threads here, the simlar threads (titles) don't always seem that relevant. Granted it has to be a work in progress so I'm happy to wait and see.
 
Is there a method of wiping the cache entirely and start with a fresh rebuild?
There's an option on the enhanced search config page to clear the similar thread cache.

I'm not totally sure why the results would appear to be that different. There was a bug fixed just before the B1 release that could've caused that. The only other way that could happen is if you have added date/forum constraints to the widget (as they don't really interact too well right now).
 
There's an option on the enhanced search config page to clear the similar thread cache.

I'm not totally sure why the results would appear to be that different. There was a bug fixed just before the B1 release that could've caused that. The only other way that could happen is if you have added date/forum constraints to the widget (as they don't really interact too well right now).
Mike, FYI follow-up...

Clearing the cache and then going back into viewing the thread again seems to have worked. đź‘Ť When installing, I didn't change of the widget options for the date range or anything else, it was literally do the XF 2.2.0 B1 beta upgrade from the ACP, poke around the forums, upgrade the addons, poke around the forums, add the widget to thread view using the default options, poke around the forums.

I'll keep an eye open, if I see the widget not matching the ACP tool checker I'll reach out in case any of you guys want to take a look.
 
Top Bottom