You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mahout.apache.org by Xiaomeng Wan <sh...@gmail.com> on 2012/01/06 20:09:04 UTC

mahout algorithm for online ad matching

Hi,
I recently got into this project involving matching online ads to
website visitors based on their profiles (pageviews, search queries,
...). So basically, learn from previous clickers and match visitor to
ad in real time. The data is large, so I am thinking about using
mahout to do it. But clicking on ads is such a rare event, I am not
sure any algorithm in mahout is proper for it.

Since it is a pretty common task, I am wondering whether anybody has
done it before and would like to share their experience, such as which
method gives better performance and which features give best results.

Any suggestions or ideas would be appreciated. Thanks in advance.

Regards,
Shawn

Re: mahout algorithm for online ad matching

Posted by Sean Owen <sr...@gmail.com>.
Sounds like a job for a recommender. If as you say the volume of ads
is relatively low, you can probably get away with the real-time
non-distributed recommenders, which is really quite easy to prototype.

On Fri, Jan 6, 2012 at 7:09 PM, Xiaomeng Wan <sh...@gmail.com> wrote:
> Hi,
> I recently got into this project involving matching online ads to
> website visitors based on their profiles (pageviews, search queries,
> ...). So basically, learn from previous clickers and match visitor to
> ad in real time. The data is large, so I am thinking about using
> mahout to do it. But clicking on ads is such a rare event, I am not
> sure any algorithm in mahout is proper for it.
>
> Since it is a pretty common task, I am wondering whether anybody has
> done it before and would like to share their experience, such as which
> method gives better performance and which features give best results.
>
> Any suggestions or ideas would be appreciated. Thanks in advance.
>
> Regards,
> Shawn