Accuracy is the share of predictions that are correct. For digit recognition, perfect performance is 100% accuracy; guessing randomly among digits 0–9 averages 10%. Accuracy and error rate sum to 100%, so maximizing one is equivalent to minimizing the other. section 2.3.2
A recommendation system has a different goal: offering people something valuable. When users rate recommendations, a system might optimize for high ratings, but users may not provide those signals. Prediction accuracy alone does not capture recommendation value. section 2.3.2
















































