Currently, matches against the node title are implicitly treated with the same significance as matches against node bodies. I think most would agree that node titles should have more weight in determining relevancy, and I'd like to look at ways to accomplish that. This can be accomplished either directly in the query or in the sort handler.

Comments

g-h-b’s picture

g-h-b commented

I think most would agree that node titles should have more weight in determining relevancy

I'm one of them. Would it be possible, to integrate this feature into Version 1.2?
I don't believe in views, therefore I use v 1.2 of similar entries - and it's working very well.

Thanks for this great module.

Gerhard

jordojuice’s picture

jordojuice commented

>> don't believe in views
Well, considering that I think this could be the case for many users I may rethink my policy going forward. The module was essentially going to be replaced with the views integrated version, but I think it would make sense and be beneficial to enough users to maintain a views version and a non-views version (2.x and 1.x respectively). So, I think we can certainly continue to support both versions and maybe even port 1.x to D7.

I will definitely be looking at doing this, but it will be tricky since I think it will mostly need to be accomplished in the query. I'll have to play with it, do some testing, and check the relevance scores to see if it can be accurate but definitely look for this in a next release.

g-h-b’s picture

g-h-b commented

There are some people around who set up a Web 2.0 Site on CHEAP webspace. I think it is a good practise then to to resign from building on these well known but heavy loaded modules.

It is possible to create interactive websites WITHOUT cck, panels, views, rules and ctools - but with a fistful of ingenious small modules - and similar entries is one of them. This almost finished german Website is made with lightweight modules only. Similar terms is placed below the content (not on front).

Thank you for maintaining similar.

jordojuice’s picture

jordojuice commented

I've been looking into this, and it looks completely plausible. I just have to index the node title separately and multiply it's weight to make it more relevant. I'll have to do some query tests tonight. The question is, how much more relevant should a node title be? Twice as relevant? Five times as relevant? Hard to say. Maybe it should be a configurable option. That can be done.

g-h-b’s picture

g-h-b commented

Configurable would be perfect.

Otherwise a title should be twice as relevant at the most - in terms of 'title vs. body'. For me it would be perfect, when the title is about 1.5 times stronger than the body-text as a whole.

But I assume that the relevance programming depends on the values of the words, whereas the field-type itself is not part of the calculation. In this case it is indeed hard to determine the 'weight' of the words in the title. For sure it depends on the length of the text too. So even 10 times as relevant could be a right value.

May I draw your attention to this issue? http://drupal.org/node/1255606

Thanks - and please excuse my English, I'm not a native speaker.

jordojuice’s picture

jordojuice commented
Status: Active » Fixed

#1: FYI, Similar Entries 1.x has been ported to D7 now.

Also, I have changed all queries in new branches to BOOLEAN queries which allow us to give more weight to words found in node titles with the > operator. So, this is fixed.

Status: Fixed » Closed (fixed)

Automatically closed -- issue fixed for 2 weeks with no activity.