I read the notes in the below links. I have a couple of questions: 1. Do I have to create MyDataModel.java, MyRecommender.java and mimic the java code in the examples/grouplens directory? Or I just put all the suggested code (from the below links in examples section) in one single .java file?
2. How do I save item similarity matrix? 3. How do I save and access recommendations for all customers ID? 4. What if I want to test recommendations with Pearson correlation metric with other metrics, e.g. cosine or modified cosine transform, tanimoto coeffcient (which is just intersection set divided by union set). Where do I write these distance metrics, amnd how do I tell Recommender to use this new metric? As you can see, I am utterly ignorant of the basics. I need some help to understand this thing. Thank you, Nagu On Sun, May 24, 2009 at 6:15 PM, Otis Gospodnetic < [email protected]> wrote: > > Nagu, > > This should more than get you going: > http://lucene.apache.org/mahout/taste.html > > And half way down the page: > http://lucene.apache.org/mahout/taste.html#Item-based+Recommender > > Otis > -- > Sematext -- http://sematext.com/ -- Lucene - Solr - Nutch > > > > ----- Original Message ---- > > From: Nagu <[email protected]> > > To: [email protected] > > Sent: Sunday, May 24, 2009 7:48:51 PM > > Subject: Item-based Collaborative Filtering > > > > Hi, > > > > I need some guidance in implementing item-based Collaborative Filtering > in > > Mahout. > > > > To give an example, I built a recommendation engine using python (and I > > don't know damn about programming in general) last year based on some > real > > customer data from my company (e.g. customers who bought this stuff also > > bought these...). I have some SQL procedures that spits the data for the > CF > > algorithm, and the python program crunches the dataset, and spits out the > > recommendations for each customer (up to 50 recommended items) and it > saves > > recommendations in a database. I created a simple web framework using > django > > to present the recommendations given a customer ID. So sales teams can go > to > > an intranet page and get recommendations for any given customers. I > update > > the whole recommendations output every 15 days. > > > > I want to produce something like this using Mahout just to get a feel of > > Mahout. It will be something like, take this customer purchase history, > and > > run the item-based CF algorithm, give me the recommendations for a given > > customer and save it in a database for me. > > > > Where can I find some step by step implementation of some examples in > > Mahout. I want to understand how this whole thing works and I want to > start > > tinkering with some real time data from the company where I work. I also > > want to build some abstraction into this machine learning so that I can > use > > the output that comes out of Mahout and feed to internal > business/customer > > process and apply some business logic on top of this to make the results > > more meaningful. > > > > I am not sure if I am asking for too much, but I think I definitely need > > some guidance. > > > > Thank you, > > Nagu > >
