Recommendation system based on user liking in mahout

How we can run a recommendation system on Apache Mahout based on user liking or browsing history? I开发者_JS百科n short on a content based websites 95% traffic by non logged in users and they will come via search engine. They only way we can unique them by using IP. Is there anyway on Apache Mahout where we can find out the similar browsing behavior of users and recommend relevant content?

Collaborative filtering should still be useful if the user "likes" (clicks a link, for instance) more than a single item (link) in your website. Also consider placing a cookie so that you might later identify the user again.

So more explicitly, if you use the IP as a user ID and give your content (let's assume articles) an item ID, you can just use a GenericUserBasedRecommender. You can treat link clicks as "likes". There's a good article containing data about more complicated systems, but also touching on the same issue as you have with non-ID'd users: Google news scalable recommendation framework

Furthermore, consider clustering the items based on content. Then you can recommend similar items without knowing the user at all.