English: The Discovery Department uses the zero results rate -- the proportion of searches that yield zero results -- to measure the performance of our search system. However, little is known about possible patterns that affect the quantity (and quality) of results our users see. In this report, we use random forest and logistic regression models to shed light on the types of queries that tend to yield zero results.

Namely, we found that whether the query has an even number of double quotes is one of the most important indicators of whether it will yield zero results. Other notable features that impact the quantity of results include: whether the query is only punctuation and spaces, whether it ends with ?, and whether it has logical operators. For a full list of features and their importance and impact, please see Figures 4 and 5.

Going forward, we may want to rewrite queries to not have quotes when the original query (with double quotation marks) yielded 0 results. We may also want to address question queries (that end with ?).