Marketplace Query Detection Using Causal Low-Inventory Modeling

The patent introduces a causal inference model as an intermediary between the search system and the identification process. This model uses propensity scores to mediate the relationship between search queries and outcomes, allowing accurate identification of under-served queries while accounting for user intent and session complexity without requiring direct observation of confounding factors

The patent replaces traditional mechanical counting methods with a statistical modeling approach. Instead of simply counting search results, the system uses causal inference models, propensity score matching, and predictive modeling to identify under-served queries, substituting a complex statistical mechanism for a simple counting mechanism to achieve higher precision

The patent extracts only the essential elements needed for causal inference from complete search session data. Instead of processing all detailed user interactions, the system extracts key variables such as query characteristics, search result counts, and conversion outcomes, thereby maintaining data completeness for analysis while reducing processing complexity

The patent performs preliminary aggregation and filtering of search session data before detailed analysis. By pre-processing data to identify relevant patterns and characteristics upfront, the system reduces the complexity of subsequent causal inference operations while preserving the information needed for accurate identification of under-served queries