Anytime Ranking on Document-Ordered Indexes

Inverted indexes continue to be a mainstay of text search engines, allowing efficient querying large document collections. While there are number possible organizations, document-ordered the most common, since they amenable various query types, support index updates, and allow for dynamic pruning operations. One disadvantage with is that high-scoring documents can distributed across identifier space, meaning traversal algorithms terminate early might put effectiveness at risk. The alternative impact-ordered indexes, which primarily top- disjunctions but also anytime processing, where terminated any time, quality improving as processing latency increases. Anytime used effectively reduce high-percentile tail essential operational scenarios in service level agreement (SLA) imposes response time requirements. In this work, we show how organized such queried an fashion, enabling strict control effective termination. Our experiments topical segments selected by simple score estimator outperforms existing algorithms, allows runtimes accurately limited comply SLA