Ad Index Pre-filtering for Search Query Matching

The patent applies preliminary action by pre-processing advertisement entries into an inverted index structure before the actual search query arrives. The index is built in advance with advertisement phrases, their components, and associated metadata organized for rapid access. This allows the delivery engine to skip the time-consuming task of sorting through all advertisements during query processing, instead directly querying the pre-organized index for relevant matches.

The patent segments the advertisement database into an inverted index structure that divides advertisements into component phrases and words. Each advertisement is broken down into its constituent phrases, which are further segmented into individual words, creating a hierarchical organization. This segmentation allows the system to quickly filter and match queries against specific components rather than examining entire advertisements sequentially.

The patent extracts and separates the indexing function from the query processing function. The inverted index is extracted as a standalone pre-computed structure that contains only the essential matching information (phrases, words, and their positions in advertisements). This extraction allows the query processing to focus solely on pattern matching against the index without re-processing the entire advertisement database, significantly reducing computational resources while maintaining reliable matching.

The system performs preliminary indexing action by pre-processing all advertisements into the inverted index structure before any queries are processed. This preliminary action includes parsing advertisements into phrases, extracting individual words, and organizing them in the index with their corresponding advertisement identifiers. When a query arrives, the system only needs to perform lightweight matching operations on the pre-processed index rather than examining all original advertisements.

The patent segments the matching process into distinct stages: first matching against the inverted index using simple phrase and word comparisons, then applying additional filtering criteria only to the subset of advertisements that pass the initial index match. This segmentation of the matching process allows detailed filtering criteria to be applied efficiently to a reduced set of candidates rather than to all advertisements, maintaining precision while reducing overall complexity.

The system performs preliminary filtering through the inverted index match before applying more complex filtering criteria. The index matching stage quickly eliminates advertisements that do not contain the query phrases or words, creating a reduced candidate set. Only this smaller subset then undergoes the more complex detailed filtering, which significantly reduces the computational burden of applying precise matching criteria while maintaining high relevance.