incubator-couchdb-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jeroen van Dijk <jeroentjevand...@gmail.com>
Subject Help with an advanced view to build a recommender?
Date Mon, 11 Apr 2011 20:10:27 GMT
Hi all,

The last couple of days I have been trying to build a view that would act as
a recommender. I have tried all the stuff that I could find/think of, but I
can't find a solution. I hope someone can tell me how I can do it or maybe
tell me that it is just not possible with one map reduce. Below is the
problem description, I hope it is clear enough.

The basic idea is to use co-occurrences of apps attached to a user to
calculate the similarity between apps. This is how the two types of
documents; users and apps, look like:

{ _id: 'user-1', _type: 'user', app_ids: 'app-1', 'app-2' }

{ _id: 'app-1', _type: 'app', user_ids: 'user-1', 'user-2' }

I was hoping the map reduce approach below would work when adding the
include_docs=true option. Unfortunately this doesn't work with a reduce
function. So the remaining problem so far seems to be to obtain the total
app count together with the co-occurrence counts.

  //map
  function(doc) {
    if(doc.type == "user") {
      var app_count = doc.app_ids.length;
      for(var i = 0; i < app_count; i++) {
        for(var j = i + 1; j < app_count; j++) {
          emit([doc.app_ids[j], doc.app_ids[i]], [0, 1, 0, {_id:
doc.app_ids[i]}]);
          emit([doc.app_ids[i], doc.app_ids[j]], [0, 1, 0, {_id:
doc.app_ids[j]}]);
        }
      }
    }
  }

  //reduce
  function(keys, values, rereduce) {
    //output is [similarity, number of co-occurrences, total number, doc]
    var output = [0, 0, 0, null];

    values.forEach(function(pair) {
      output[1] += pair[1];
      output[2] = pair[2].user_ids.length;
      output[0] = output[1] / output[2];
      output[3] = pair[3];
    });

    return output;
  }


Hopefully someone has new insights that can help me a bit further. Thanks.

Jeroen

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message