When someone searches an online store for 4K televisions, recall asks: how many of all relevant televisions did the search actually return? It measures retrieved relevant documents against the complete set of relevant documents. A result set with high recall omits fewer useful matches, although it may include more irrelevant results. section 8.1 section 10.1.2
A 4K television search
Suppose nine documents are relevant to the query and six are returned. Three relevant televisions are omitted, so they are false negatives. Recall = 6 / (6 + 3) × 100% = 66.6%. The search found two-thirds of the relevant documents. section 10.1.2
Recall works alongside precision, which favors the relevance of returned results. These measures tend to oppose each other: pushing for more complete retrieval can admit less-relevant documents, while selecting only the best matches can omit useful ones. section 8.1 section 10.1.2
Use recall when evaluating or tuning search:
- Inspect omitted relevant documents as false negatives. section 10.1.2
- Adjust match queries, filters, and boost to explore the balance between recall and precision. section 10.1.2
Recall = 6 / (6+3) × 100% = 66.6%
Use the compact formula to follow the television-search calculation. section 10.1.2
