PDA

View Full Version : Rank search results by relevance


xja
06-27-2005, 10:08 PM
First of all, many thanks to Kinook for the improvements in 1.3b. UR just keeps getting better thanks to your incredible responsiveness to customer feedback.

To add another request or two on the pile:

It would be great if the search results could optionally be ranked by relevance, like Internet search engines do (based on how often a keyword occurs, if they occur in the title, if they are in proximity to each other, etc.). I am sure that is not an easy one to implement, but it would be helpful.

Also, how about a Quick Search box (where you could enter the search term) on the toolbar?

quant
07-06-2007, 06:53 AM
I'd like to revive this two years old thread (started by xja, another one that never got implemented ;-) ).

could you please reconsider adding relevance (system attribute) column to search results? Thank you

quant
10-19-2007, 05:04 AM
I use UR on a daily basis, and even though I put inside my main database only the most important info, I start to have problems when inserting new information to keep my db consistent and without repetitions, or simply when trying to search for sth.

The problem is to find to most relevant data based on your search. With a big db, it is simply not enough just to return items with given keywords. For example I try find whether I already have sth similar in the database. Say I'm going to insert sth about Riemann integrable functions, I can try to search for these three keywords, but the search result might return nothing, so I can try to search for Riemann integrable or Riemann functions or integrable functions ... maybe again nothing or too many results ... if there are too many results, it's hard to see which are the most relevant because there is no ranking.

The point is, that the relevance search would be the most appropriate for this, and would help a lot in the process of keeping db "healthy".

The basic "relevance" attribute could simply be based on

1. how many of the searched keywords it has,
2. how often a keyword occurs,
3. and if they occur in the title.

The point 1 is very important, cause it doesn't need to have all of the searched keywords and it is very time consuming or impossible to create a search with all the combinations of the searched keywords.

There are many PIM programs that allow fast insertion of various informations, but the real strength of any of such programs shows only when you want to get sth out of it. The fast indexed search is great, but nothing special by today's standards.

IMHO, the relevance search is the next step in the direction of making a great PIM program.

Thank you! Any ideas?

quant
10-19-2007, 05:20 AM
This should be very easily implemented cause the search goes through the index anyway, it could be a first (system) calculated attribute in UR that would be automatically updated when the search is performed.

My suggestion for simple formula for relevance attribute:

20*(number of found (unique) keywords out of all searched)
+ 10*(number of searched (unique) keywords in the title)
+ 1*(sum of occurrences of all found keywords).

The explanation is simple:
If the item has certain keyword 10 times, it is the same as if it was just once, but in the title. And it would need to have say 20 of occurrences of some keywords to be at least as relevant as another item that contains one more (but unique) keyword. Or it would need to have two keywords in the title to be as relevant as another item that has one more unique keyword ...

what do you think?

quant
06-17-2008, 04:48 PM
seems like I'm the only one who misses this feature.

Today I found (suggested by user at http://www.outlinersoftware.com) an excellent file launcher that allows great customization for scoring search result (heuristic scoring, pattern scoring). While the search often results in many hits, by customizing the program to your needs you get what you look for much faster with less effort (isn't this one of the main points of using UR?).

I still cannot use UR for a management of big amount of data because other programs do SEARCHING much much better.

For example, while UR finds keyword in a pdf file, it doesn't count the number of occurrences, or the proximity of more than one searched terms. Any decent search program will put more score to more occurrences of the search terms, etc ...

pleeeeeeeeeeeeease

dasymington
06-23-2008, 04:52 PM
This is one of the features that I really miss since moving from Asksam to UR.

kinook
10-14-2009, 06:20 PM
This is available in v4.1.

quant
10-14-2009, 06:35 PM
Originally posted by kinook
This is available in v4.1.
testing ...

I'm going to compare with Archivarius3000 which is my main search engine for some time, will get back with results.