PredictionIO suggest to like items that have already been liked

Question

I'm trying to use PredictionIO recommendation engine in Rails app to suggest items for users to like. So, I have three models: user, product and favorite(user_id, product_id). This is what algorithms.json file looks like:

[
  {
    "name": "ncMahoutItemBased",
     "params": {
      "booleanData": true,
      "itemSimilarity": "LogLikelihoodSimilarity",
      "weighted": false,
      "threshold": 0.6,
      "nearestN": 10,
      "unseenOnly": false,
      "freshness" : 0,
      "freshnessTimeUnit" : 86400
    }
  }
]

The things is, after training and deploying, I get a list of suggested items for user and some of which the user has already liked. Why is this?

What is the name for UserBased algorithm instead of "ncMahoutItemBased"?

Thanks.

Answer 1

There is nothing wrong with recommending an item the user has shown a preference for. This is expected behavior in a clothing store, where I always buy Levi's Jeans and they want to remind me of that.

In your case you may not want to recommend items already prefered so filter them out of the recommendations. In most Mahout recommenders this is done for you so PredictionIO must have disabled that feature. Is there some param or config option that tells PredictionIO to filter out a user's preferred items?