Precision and Recall

Synonyms

Positive predictive value; Sensitivity; False negative rate

Definition

Recall measures the ability of a search engine or retrieval system to locate relevant material in its index. Precision measures its ability to not rank nonrelevant material. With everything above rank cut-off n considered “retrieved” and everything below considered “not retrieved,” precision and recall can be stated mathematically as:

$$\eqalign{ {\rm precision}\, = \,{{\left| {{\rm retrieved}\,\& \,{\rm relevant}\,{\rm at}\,{\rm rank}\,n} \right|} \over {\left| {{\rm retrieved}\,{\rm at}\,{\rm rank}\,n} \right|}} \cr {\rm recall}\, = {{\left| {{\rm retrieved}\,\& \,{\rm relevant}\,{\rm at}\,{\rm rank}\,n} \right|} \over {\left| {{\rm relevant}} \right|}}}$$

Key Points

Precision and recall are the traditional metrics for retrieval system performance evaluation, and nearly all other performance measures can be seen as either precision-based, recall-based, or a combination of the two.

Table 1