Prediction quality values

The "quality" value refers to the degree a user will like an item. The LikeMindsserver presents predictions in decreasing order of quality.

Recommendation quality is calculated using this equation:

quality = confidence^mpowx (pred rating —mean pred rating) + K x confidence^pow

...where...

• confidence is the confidence value for the recommended item.
• pred rating is the predicted rating for the item.
• mean pred rating is the mean recommended rating for all items for the user.

To specify ^mpow, set db.applic.param.BestBets.MultPower:

db.applic.param.BestBets.MultPower = 0.5

To specify K, set db.applic.param.BestBests.Coefficient:

db.applic.param.BestBets.Coefficient = 0

To specify pow, set db.applic.param.BestBets.Power:

db.applic.param.BestBets.Power = 1.0

We can use these parameters for all of the Recommendation engines except for the Item Affinity Engine.

Parent: Specify recommendation behavior