Posted by laura s on March 21, 2011 at 2:59pm
Once upon a time there was a Content Recommendation Engine that showed a lot of promise. http://drupal.org/project/cre It never got beyond Drupal 5.
The concept, to provide recommendations of content based not just on the content but on the user's own usage metrics, is in much use on the web, but as far as I know there is no viable solution currently in Drupal.
A Content Recommendation Engine update to Drupal 7 would be awesome, and something that could be used by almost any Drupal site that has active members and readers.
I would be happy to volunteer as a mentor on user behavior algorithms and user interface design. A D7 code mentor would be needed as well, imho.
Comments
CRE and scalability?
I like the idea, though I've got my doubts on the scalability of the approach taken by CRE. Recommendation algorithms are on general very compute intensive. Certainly when the number of users and content is high, these calculations rapidly take a long time to finish, and using php makes it only worse. I think that to make this work, you should do something similar as the Solr module, use an external system to do the calculations. In fact that's the approach that the recommender module ( http://drupal.org/project/recommender ) is going to take ( see the roadmap here : http://michiza.com/blog/roadmap-recommender-api-module-and-helper-modules ) . Up until now not much work has been done to achieve this, so maybe this would a good opportunity to get this rolling? In any case I agree with you that drupal would benefit from a great recommender engine.
the recommender API module
Thanks jolos. I'm the developer for the recommender module (http://drupal.org/project/recommender). I'll submit an application soon. Hopefully I can get RecommenderAPI integrate with Mahout this summer. For performance/scalability reasons, there's just no way we can implement recommendation algorithms in PHP.
Please refer to the proposal
Please refer to the proposal at http://groups.drupal.org/node/137054 :)
This is a good proposal that
This is a good proposal that has a lot of support. Seems like a solid way to go. :)
PINGV | Strategy • Design • Drupal Development
Tracking User behaviour
Hello,
We can store all the keywords associated with the "current" user's browsing pattern of different articles on the website in a cookie and then we can recommend him/ her new pages/ articles based on this keyword listing. We can truncate the old keywords from the cookie if it grows large enough.
As for the privacy concerns, we will store the cookie only on the user's browser and just use the data for recommending appropriate articles. We don't need to store it anywhere on our DB.
For new users, for whom there is no browsing history available, we can recommend them the most popular yet from our data.
As for the large websites, which have large demographics reach, we can allow the site admin to "recommend" articles from the user's demographics perspective. As in, recommending him article based on browsing pattern of another user from the same demographics region. (This will involve storing users' browsing pattern but can be done anonymously & with consent, if needed)
As per the discussion with "agentrickard" we will make this with APIs so other modules can benefit from it (rather than making it a module itself). They can then use any of the above (more suggestions welcome :) )with default one as the "related content (by keywords)". This approach won't be much processing complex and would be needing low resources.
Please let me know your views about it. I'm willing to take this as GSoC project this summers.
Thanks,
Neeraj Agarwal