Evaluating recommendations

Understanding how good a collaborative filtering system is can be broadly determined by measuring three types of accuracy parameters, namely:

• Prediction Accuracy Metrics

  These measures help to understand how accurately the recommender works. These measures work by calculating the differences between previously rated items and their ratings estimated by the recommender system.

• Decision Support Metrics (a.k.a Confusion Matrix)

  These measures are used to find how well a supervised learning algorithm has performed.

• Ranking Accuracy Metrics

  These metrics are used to find out how well the recommender has placed the items in the final recommended list.

Prediction accuracy

Metrics help us to understand how good the predicted ratings are. Here are some of the prediction accuracy metrics that are used frequently:

Note

denotes the predicted rating for user on item . And is the actual rating. So the closer the value of these metrics to zero, the better the prediction algorithm...