Query Intent Segmentation for E-commerce Search Relevance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
E-commerce sites face challenges in processing user queries efficiently, leading to irrelevant search results and potential loss of sales, as existing methods fail to accurately categorize user intents and filter results effectively.
Innovation Solution
The system improves query processing by analyzing both historical and predetermined queries to identify core and ancillary portions, using mappings to filter results to the most relevant category, and combining these analyses to enhance query categorization, thereby providing more targeted and relevant search results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If exact match mapping of queries to categories is used, then query processing speed is improved, but query result relevance deteriorates when users add additional words or tokens
Solution Approach 1:
The query is segmented into a first query portion and a second query portion. The first portion is used for exact match mapping to categories to ensure fast processing, while the second portion contains additional words or tokens that are excluded from the mapping process. This segmentation allows the system to maintain speed while improving relevance by focusing the category mapping on the core intent portion of the query.
Solution Approach 2:
The system extracts and excludes the second query portion (additional words or tokens) from the query mapping process. By taking out these non-essential portions that would interfere with exact matching, the system prevents them from causing mismatched category assignments while still processing them to provide comprehensive search results.
2Measurement precision
If historical query analysis is performed to improve categorization accuracy, then query result relevance is improved, but system complexity increases
Solution Approach 1:
The system performs preliminary analysis of historical queries to identify patterns and relationships between different query portions and their corresponding categories. This pre-computed knowledge is stored and reused during actual query processing, improving categorization accuracy without requiring complex real-time analysis for each query.
Solution Approach 2:
The system introduces an intermediary mapping structure that connects query portions to categories based on historical analysis. This intermediary layer simplifies the complexity by providing a pre-established relationship map between query patterns and categories, avoiding the need for complex algorithms during actual query processing.
3Quantity of substance
If all query results are returned without filtering, then result completeness is improved, but user effort to find relevant items increases
Solution Approach 1:
The system applies different quality levels to different portions of the query results. Results matching the first query portion (core intent) are provided with higher priority and more prominent display, while results matching the second query portion (additional words) are included but with lower priority. This local quality differentiation helps users quickly identify the most relevant items without hiding less relevant results.
Solution Approach 2:
The system performs partial filtering by focusing primarily on the first query portion for category-based result selection, while still including results related to the second query portion. This partial action approach ensures that the majority of results are highly relevant to the core intent, reducing user effort, while maintaining sufficient completeness by not completely excluding results related to additional query words.
Data Source
AI summary
Methods and systems for improved categorization of queries are disclosed. In one aspect, a search query is received and determined to match a pattern having an uncategorized variable placeholder portion. In response to determining the search query matches the pattern, the pattern is used to identify a first portion of the search query based on the first portion of the search query corresponding to the uncategorized variable placeholder portion of the pattern. The search query is categorized by assigning, to the search query, a category associated with the first portion of the search query. Search results are generated for the search query using the category associated with the first portion of the search query.