There are a few problems that you have. 1) user-based recommendation is often slower than item-based (sometimes MUCH slower). This can make a 2-10x difference in practice
2) pre-computing recommendations is usually much less efficient than computing them on the fly (because typically few users will require their recommendations before you recompute them). This can make 1000x difference in practice. 3) pre-computing recommendations is almost always less accurate than computing them on the fly because you don't have all of the recent user history when you pre-compute the recommendations. 4) pre-computing recommendations is also a problem if you have new users since the last bulk computation. If you are using item-based cooccurrence recommendation, however, you can still compute recommendations for these new users almost as soon as they have done anything on the site. If you just run itemsimilarity and store the results in a search index in order to provide real-time results, I think that you are likely to be much more happy. On Fri, Nov 20, 2015 at 4:45 PM, Gughan Raj <gughan....@indiaproperty.com> wrote: > Hi, > > I have been using mahout for running a user based recommendation. My data > has 1M users and 3M associations. I want to write the output of the > recommender object to a file and using a long primitive iterator takes lot > of time. I want to know is there a way to write all recommendations to a > file efficiently. > > -- > Regards, > Gughan Raj S | +91-9500022771 >
