On the demo Solr+Mahout recommender site there are recs made from your most resent detail views. These are calculated using the user’s most recent views as the query against the training data, which comes from stronger preferences (ratings). In an ecom case it would be like views for the query against a model built from purchases.
Strictly speaking the recent-view recommendations should be cross-recommendations. They are purchase recs made from views. As enough view data is built up we’ll switch to cross-recs for these and see if the quality improves. I suspect there are going to be more and more uses for the cross-recommender as people apply the idea.
