Overview

This document summarizes how search queries are processed within the Hawk Search Platform.

Note: The process outlined assumes that all features are being utilized and does not account for client-specific customizations which may impact the processing order.

Search Query Process

When keywords are sent to the platform to perform a search query, these are the steps performed:

1. Retrieve User Profile
   To utilize features such as Visitor Targets, a User Profile is retrieved. Geolocation data based on visitor’s IP and other custom profile information are saved.
2. Load and Cache Global and Conditional Boost Rules
   All active Workbench rules related to adjusting relevance ranking are retrieved.
3. Load and Cache Personalization Insights
   Insights from the Personalization Module are retrieved to adjust relevance ranking. These insights enable 1-to-1 Personalization in search results.
4. Load and Cache Visibility Rules
   Visibility Rules in the Workbench filter which records would be excluded from the results.
5. Load and Apply Measurement Conversion
   Measurement Conversion is applied for keywords.
6. Perform Search
   1. Load and Cache Stop Words: Words that should be ignored are retrieved.
   2. Load and Cache Synonyms: The Synonym list from Workbench is loaded.
   3. Load and Cache Spell Check Override: The Spell heck Override list from Workbench is loaded.
   4. Load and Cache Learning Intelligence: Intelligence from the Learning Search feature is loaded.
   5. Execute Search
      1. Tokenization: The Engine analyzes the query string and yields a sequence of distinct terms.
      2. Apply Scoring: For detailed technical documentation regarding scoring, please read: https://lucene.apache.org/core/3_5_0/scoring.html
      3. Apply Boost Rules, Personalization Insights, and Visibility Rules: Relevant rules and insights are applied.
      4. Facet Filtering: Selected facets are applied to the results.
   6. Retrieve Spell Correction / Did You Mean Options: Relevant spell correction and Did You Mean options are retrieved.
   7. Load Best Fragment for Each Search Result Row
   8. Load Facets: Relevant facets based on the search results are retrieved.
7. Process Did You Mean
   Alternative search terms will be returned if applicable.
8. Process Redirect Rules
   If the input triggers a Redirect Rule, the value will be returned.
9. Process Merchandising Rules
   Merchandising Rules that are triggered and relevant for the search query are returned.
10. Process Tracking Data
    Tracking data influences features such as Learning Search and Personalization.