How to find trending tags out of long search string

views:

answers:

+2 Q:

How to find trending tags out of long search string

Hello,

I would like to have a trending tags in my website according to searches users made. The problem I'm not seeing an simple solution for is how to easily extract the important terms out of a search string. For example, many users might search for "visual studio" with different purposes. For example, "visual studio 2010", "visual studio unit testing", "visual studio web forms components". In those 3 searches, "visual studio" is trending. How can an algorithm notice that since "visual studio" in most cases will be mixed with many other words?

Thank you!

+1 A:

Have a look on this codeplex project

http://www.codeplex.com/TheTagCloud

Includes a function that you can pass an html file to as input and will return a tag cloud.

Chris Ballance 2009-10-27 16:17:12

TheTagCloud seems to be based on the tags already being known. I think this question is trying to find out how best to build the tag set from the content.

Stephen Doyle 2009-10-27 16:22:45

@Stephen The example only shows creation from a list of strings. The project supports an html page input also.

Chris Ballance 2009-10-27 16:26:10

Yes, I've actually written something like TheTagCloud. What's being harder to find is how to extract precise tags based on searches made on a website.

JP Araujo 2009-10-27 17:36:28

+2 A:

split every search query into an array of single words.
calculate the distance between the words (the nearer, the better => higher value)
add this word distance for each wordpair across all queries.

The wordpairs with the higher values are your "trending tags".

Martin Hohenberg 2009-10-27 16:19:01

That sounds interesting. Would you give more details on this? The idea is to compare all possibilities or only a word with the the others that come after it, up to the end of the array?

JP Araujo 2009-10-27 17:34:26

I used to do this on a many-to-many way: compare any two words within a given string. This makes sense once you realize that "prices visual studio" also uses visual studio as a "trending tag". On the other hand, you could also do this only regarding the firstWord to nth-word relationships (which saves computing time, but disregards "trailing tags" (seriously, there *must* be a better term ;) ) that appear later in the string.

Martin Hohenberg 2009-10-27 18:44:28

ansaurus

tags:

views:

answers:

How to find trending tags out of long search string

related questions