The Mahout recommendation code is not a program. It is a modular toolkit. There are 5 different kinds of tools, and a few implementations of all 5 kinds. This page describes the toolkit design: https://cwiki.apache.org/confluence/display/MAHOUT/Recommender+Documentation
Your best avenue to learning and using these tools is to buy the "Mahout In Action" book, published by Manning. It is being written by some of the core Mahout authors, and is very helpful. Download PDFs are available for most of the book, but it is not in print yet: http://www.manning.com/owen/ On Tue, Feb 22, 2011 at 12:40 AM, Sean Owen <[email protected]> wrote: > It does both. It's going to compute in real-time, unless you > pre-compute by using a GenericUserSimilarity to precompute from a > UserSimilarity implementation. Using a CachingUserSimilarity wrapper > might be a smarter way to go. > > Yes, this might be essential to respectable performance depending on > your algorithm, DataModel, and data size. > > 2011/2/22 王世森 <[email protected]>: >> Hi all, >> >> >> >> I`m a beginner in building recommendation system. I have a very simple >> question, when we use mahout to do a CF recommendation, for example we use >> the basic user-based recommendation, is the similarity between 2 users(or >> the neighbourhood of one user) pre-computed or computed when the specific >> recommendation is performed? >> I found when I run the examples in mahout, the recommendation seems to be >> performed in the latter way which makes it quite ineffective. Since I`m >> doing a CF recommendation project, in which millions of users is involved so >> I hope I can make the CF recommendation in a more effective way. >> >> > -- Lance Norskog [email protected]
