Voice Search Entity Resolution via Department Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice-controlled systems struggle to accurately process vague or less definite voice search queries, particularly when users provide follow-up queries with additional attributes, as they find it challenging to correctly identify the user's goal and distinguish between new searches and refinements.
Innovation Solution
The system employs a structured graph-based approach, using natural language processing and machine learning algorithms to identify product attributes and filter refinements, determining whether the input is a new search or a refinement of a previous query, and applies these filters to refine search results within the product catalog.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If voice-controlled systems use simple query processing, then the system is easy to operate, but the system cannot accurately process vague or follow-up queries with additional attributes
Solution Approach 1:
The patent introduces an entity resolution service as an intermediary component between the voice query interface and the product catalog search system. This service resolves entity mentions in voice queries to specific product attributes by comparing against the catalog schema, enabling accurate interpretation of vague or follow-up queries without requiring complex processing throughout the entire system.
Solution Approach 2:
The patent segments the query processing system into distinct modular components: voice query reception, entity resolution service, attribute matching engine, and search execution module. This segmentation allows each component to specialize in specific tasks, improving overall query understanding accuracy while maintaining manageable system complexity through clear separation of concerns.
2Reliability
If the system processes all possible search attributes, then the search results are comprehensive, but the system cannot efficiently distinguish between new searches and refinements of previous queries
Solution Approach 1:
The patent implements feedback mechanisms where the system analyzes the relationship between current voice queries and previous search results to determine whether the user intends to perform a new search or refine existing results. This feedback loop enables the system to adapt its processing behavior dynamically, improving both accuracy and efficiency by avoiding redundant processing of obvious refinements.
Solution Approach 2:
The patent makes the search processing behavior dynamic by allowing the system to switch between different operational modes (new search vs. refinement processing) based on real-time analysis of query patterns and context. This dynamic adaptation enables efficient handling of follow-up queries while maintaining comprehensive attribute processing when needed.
3Quantity of substance
If the system returns all matching products, then the search results are complete, but the results include out-of-department products that reduce user experience
Solution Approach 1:
The patent applies local quality filtering by introducing department-specific filtering rules that are applied locally to search results based on the product catalog structure. This allows the system to maintain complete search results within relevant departments while excluding out-of-department products, thereby improving user experience without sacrificing result completeness for the intended scope.
Data Source
AI summary
Voice product search requests in a product catalog are improved by identifying search refinements versus new search requests and by applying filters of departments to identify sub-sets of the product catalog to search based on the search string. The systems and methods described herein provide for analysis of voice search data, identification of departments within a product catalog that may be applicable to the search request and the search refinement, limiting the scope of the search, matching the voice input to an attribute type, and performing the refinement on the search results.