202602250206
Status: #reference
Tags: Information Retrieval
State: #nascient

Interpolated Precision

This is one of the most crucial metrics used in information retrieval systems as it gives us a way to compare a system across queries (which is not really doable for precision, Recall (Statistical Power) and F1-Score alone), but also to get a vision of the behavior of the system across recall points to potentially flag abnormal behavior.

This in English tells us:
"What is the best precision I can hope to achieve if I desire a recall of X%."

As such, it is simply a function which tells us at each recall level for a given query, what is the best precision obtained. For that reason it is non-increasing, each increase in recall level can only diminish the total pool upon which we apply our max, therefore we can only ever hope to maintain the same precision at a higher recall, never increase it.

Relevant Links