Intent-Aware Recommendation System for Search Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to accurately identify and persist user intent in search queries, leading to irrelevant recommendations and increased user interaction, as they do not formally recognize the mission behind a search query, resulting in repetitive actions and unnecessary information downloads.
Innovation Solution
A machine learning-based system that autolabels natural language search queries with electronic catalog fields to infer user intent, trains a model to predict intent from new queries, and customizes user interfaces to align with the user's mission, reducing irrelevant results and repetitive interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional recommendation systems are used that do not formally identify user intent, then the system structure is simpler, but the relevance of recommendations deteriorates and user interaction increases
Solution Approach 1:
The system segments the user search process into distinct components: intent identification, query labeling, and recommendation generation. By breaking down the complex task of understanding user intent into manageable segments (identifying mission, extracting attributes, matching with catalog fields), the system achieves higher measurement precision while keeping each component's complexity controllable
Solution Approach 2:
The patent introduces an intermediary mechanism - the query labeling system with electronic catalog fields - that bridges the gap between raw search queries and recommendation generation. This intermediary layer formally captures user intent through structured labels, enabling accurate intent identification without requiring the entire system to be overly complex
2Loss of information
If the system provides generalized recommendations without persisting user intent, then the device complexity is lower, but the loss of information about user mission increases
Solution Approach 1:
The system performs preliminary action by identifying and labeling user intent at the query stage, before recommendations are generated. By capturing the user's mission and attributes upfront through query labeling, the system preserves this information throughout the interaction session, preventing information loss without requiring continuous complex processing
Solution Approach 2:
The query labeling system serves multiple functions: it identifies user intent, structures search queries, enables recommendation generation, and persists information across interactions. This multi-functional approach reduces information loss while avoiding the need for separate complex systems for each function
3Productivity
If the system displays all search results without filtering by inferred intent, then the quantity of information provided is higher, but the relevance to user mission deteriorates
Solution Approach 1:
The system extracts only the essential attributes from user queries that are relevant to their mission (e.g., size, color, type). By taking out and isolating these key attributes through query labeling, the system filters recommendations to match user intent precisely, improving productivity by eliminating irrelevant information while maintaining appropriate information volume
4Measurement precision
If the system requires users to manually specify all search attributes, then the precision of search results is higher, but the ease of operation deteriorates
Solution Approach 1:
The system practices self-service by automatically identifying and labeling search attributes from the user's query without requiring manual specification. The query labeling mechanism extracts attributes autonomously, achieving high search attribute accuracy while maintaining ease of operation, as users simply need to enter their natural language query
Data Source
AI summary
The present disclosure is directed to training and using machine learning models to determine user intent from a search query, for example via a semantic parse that identifies particular catalog fields for items in an electronic catalog that would satisfy the user's current mission as reflected in their search query intent. The determined intent can then be used to filter recommendations and/or pre-select attribute-value input fields on detail pages displayed after the user navigates away from the search results page, until the mission is complete.