Commons comes with some blocks that guide you towards new content through "most active" groups and newly posted groups. However, these algorithms are not useful for installations with a lot of content. The two items that would greatly improve knowledge management are trending posts and recommended content. There some existing solutions and technologies that could be leveraged to provide simple but useful suggestions that would help cut through the fluff in larger document sets.

Trending posts

In any community it is important to show the trending / most popular content. The Radioactivity module provides a lot of options that helping this area. This isn't revolutionary stuff, just wanted to get the idea down.

Recommended content

Although there are next-gen recommendation engines such as Apache Mahout, it would be a low level of effort to tie in the benefits of Solr to provide personalized results and recommended content via the MoreLikeThis capability of Solr. We have some data points such as User Terms and the "My interests" area that could be used to boost certain documents in the search results and perform the MoreLikeThis queries.

Comments

What about Recommender API, seems like 'users that join this group also joined...' 'users who browsed this browsed' and 'users that flagged this' would go down quite well?

In terms of Solr MoreLikeThis, I used this once in tandem with a hidden tag vocabulary which used Alchemy/Yahoo to automatically tag terms, then used similarbyterms to provide 'Similar items'. Generally I found the MoreLikeThis results really difficult firstly to configure and then to further tame, in terms of filtering what's produced, and the similar by terms results even without taming through Views were often the same or better. Plus a solution like that could work OOTB rather than messing about installing Solr on a server.

Though don't get me wrong I'd really like to be able to use Solr for recommendations and would gladly be told forthrightly that I'm completely wrong and just had Solr configured badly... but like I say in my case it seemed like by chanelling content through Solr I got a fast recommendation engine but one that couldn't fine-tune what it threw back because it wasn't Drupal and so couldn't take advantage of the Drupalised data I'd put in it.