12

(I admit I am no expert in graph databases or NoSQL, having only used it for a few hobby projects so far.)

I've been using technologies like InfiniteGraph and Stig for recommendations - these are graph databases that supposedly are optimized for tasks like this. It looks like the new Google Predictions API is capable of serving the same purpose -- given a data set and a user's actual likes as a subset, be able to predict what the user might actually like.

Is there a sure-metric to compare Google Predictions with other graph-based databases?

ronalchn
  • 12,225
  • 10
  • 51
  • 61
ina
  • 19,167
  • 39
  • 122
  • 201
  • One obvious difference is that Google Predictions is a cloud based service while others are not. Haven't seen any comparison thus far though. – Bjoern Rennhak May 11 '13 at 22:56
  • You can train the algorithms with a subset of your data and test them with the remainder. Also, I didn't think the graph databases came with recommendation algorithms. Only they provide a mechanism for implementing your own. Unless you're talking about this (http://docs.neo4j.org/chunked/milestone/cypher-cookbook-similar-favorites.html), which is just a trivial example, IMO. – Paul Jackson Mar 23 '14 at 14:20

1 Answers1

1

The prediction is quite obvious and right. But as per my knowledge, Google Prediction API uses Page ranking mechanism; not sure about graph database. Unlike Facebook, Google might using GDB for Google+, but in one of official neo4j blog they haven't mentioned anything about Google.

ShreyansS
  • 167
  • 1
  • 2
  • 10