推荐系统

推荐系统评价指标

1. Precision 精度    intersectionSize / numRecommendedItems

2. Recall 召回率  intersectionSize / numRelevantItems

3.  Fall-out (numRecommendedItems - intersectionSize)/(numItems - numRelevantItems)

4. nDCG

  double cumulativeGain = 0.0;   

    double idealizedGain = 0.0;   

    for (int i = 0; i < recommendedItems.size(); i++) {   

      RecommendedItem item = recommendedItems.get(i);   

      double discount = i == 0 ? 1.0 : 1.0 / log2(i + 1);     

    if (relevantItemIDs.contains(item.getItemID())) {    

       cumulativeGain += discount;   

      }       

  // otherwise we're multiplying discount by relevance 0 so it doesn't do anything

        // Ideally results would be ordered with all relevant ones first, so this theoretical      

   // ideal list starts with number of relevant items equal to the total number of relevant items    

     if (i < relevantItemIDs.size()) {        

   idealizedGain += discount;       

  }      

}      

nDCG.addDatum(cumulativeGain / idealizedGain);

In computing, assume relevant IDs have relevance 1 and others 0