Retrieval Metrics

In Retrieval System there are 3 types of metrices

  1. Predictive Metrics: It shows how much of the items are really accurate / relevant
    1. For graded list
      2. Mean Absolute Error (MAE)
      3. Mean Squared Error (MSE)
      4. Root Mean Squared Error (RMSE)
    2. For binary list
      1. Precision@K
      2. Recall@K
      3. F-Beta@K
  2. Ranking Metrics: It shows how well the system could sort the retrieved items
    1. Mean Reciprocal Rank (MRR)
  3. Behavioral Metrics: It shows different specific properties of the model, like diversity, novelty
  4. Business Metrics: It shows the business profit or evaluation like user clicks, user buy, user retention, user watched and so on.

References


Related Notes