[DEPRECATED: This approach does not scale. I highly recommend looking into a graph database such as Neo4j to solve this problem.]

RecEngine

Lightweight, easily implemented recommendation engine. This package is for users who are "liking" things, "purchasing" things, or "voting" on things.

$ meteor add samcorcos:recengine

Associate items based on user "upvotes", "likes", "purchases", etc using the syntax:

recEngine.link('<USERID>', '<ITEMID>')

Then, to get a recommendation, use the syntax:

recEngine.suggest('<USERID>', <NUMBER OF SUGGESTIONS>, function(error, result) {
  // "result" is an array of objects
  // Within the "result" object is "suggestion" and "weight"
  // "suggestion" is the item being suggested
  // "weight" is the strength of the suggestion -- higher number means stronger suggestion
})

Algorithm

Suggestions are produced using the Ford-Fulkerson algorithm, which computes the maximum flow in a flow network.

To provide an analogy, think of the algorithm as finding the most efficient way to drive from point A to point B, taking into consideration the number of lanes each street has—streets with more lanes are better.

Much of the logic comes from: https://gist.github.com/methodin/1561824

Runs in O(E*Fm) time, where E is the number of edges, and Fm is the maximum flow.

Basic Example

This works not only with USERIDs, but with colloquial names and just about anything else. For example, you could install the package and write the following:

if (Meteor.isServer) {
  Meteor.startup(function () {
    recEngine.link("Mike", "cake");
    recEngine.link("Mike", "coffee");
    recEngine.link("Mike", "pie");
    recEngine.link("Sarah", "coffee");
    recEngine.link("Sarah", "cake");
    recEngine.link("Alex", "yogurt");
    recEngine.link("Alex", "cake");
    recEngine.link("John", "cake");
    recEngine.link("John", "coffee");
    recEngine.link("John", "pie");
    recEngine.link("Nick", "coffee");
    recEngine.link("Nick", "cake");
    recEngine.link("Sally", "yogurt");
    recEngine.link("Sally", "cake");
    recEngine.link("Zeke", "cake");

    recEngine.suggest("Zeke", 2, function(err,res) {
      if (err) console.log(err);
      console.log(res);
    })
  });
}

... and your server console should read:

[ { suggestion: 'coffee', weight: 3 },
  { suggestion: 'pie', weight: 2 } ]

This means that Zeke would most probably like coffee given the preferences of his peers, and would probably also like pie, though, to a lesser degree.

More Practical Example

Say you have an app that keeps track of "likes" or "upvotes". In the method you wrote to keep track of upvotes, you can add a call to recEngine.link('<USERID>', '<ITEMID>') in your function. For example:

Meteor.methods({
  upvote: function(movie) {
    var userId = Meteor.user()._id;
    Movies.update({ _id: movie._id}, {$addToSet: { upvotes: userId} }, function(err, res) {
    if (err) console.log(err);
    recEngine.link(userId, movie._id);
    });
  }
})

Then, when you have enough upvotes in your database (it doesn't take very many), you can make suggestions in the following format:

Meteor.methods({
  suggest: function(numSuggestions) {
    var userId = Meteor.user()._id;
    recEngine.suggest(userId, numSuggestions, function(err, res) {
      if (err) console.log(err);
      return res;
    })
  }
})

This will give you how ever many suggestions you declared as the variable numSuggestions (could be any number) for the currently logged in user.

Additional information

Keep in mind that this package only allows users to vote on each item once, which works for most voting systems, but not all.

To Do

Optimize performance by storing suggestions
Configurable number of records stored, to allow developer to change performance recEngine.config.numRecords
Add a method for finding similar users recEngine.similarUsers
Add a method for finding similar items recEngine.similarItems
Divide inputs into contexts so retrieval can be categorized

samcorcos / recEngine