Session-Based Query Refinement for Search Result Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current networked marketplaces face challenges in accurately determining user search results due to generic search terms, presentation bias, and insufficient data from sales, leading to incomplete and inaccurate product recommendations.
Innovation Solution
Implementing a system that uses session-based refinements, where a query module receives user queries and a refinement module creates associations between initial queries and user actions, such as item selections and purchases, to generate accurate search results by building a hierarchy of query refinements and scoring refinements based on user events, allowing for improved search result accuracy even with limited search parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the system presents a variety of products including generic search terms, then the system can accommodate diverse user intentions, but the system suffers from presentation bias and cannot accurately determine user preferences
Solution Approach 1:
The patent segments search results into different categories based on user interactions and refines queries by dividing the search space into more specific product subsets. The system tracks which product categories users view and purchase to segment the generic search results into more precise recommendations, thereby reducing presentation bias while maintaining versatility.
Solution Approach 2:
The system dynamically adjusts search results based on real-time user interactions during a shopping session. The query refinement mechanism continuously updates product recommendations based on user views, selections, and purchases, making the search results adaptive to individual user intentions rather than static, which improves accuracy without sacrificing the ability to accommodate diverse users.
2Reliability
If the system relies on sales data to determine user preferences, then the system can identify purchase patterns, but sales represent a very small data sample compared with item views or selections
Solution Approach 1:
The system performs preliminary actions by tracking and recording user interactions (views, selections) before purchases occur. These preliminary interaction data are stored and analyzed to build predictive models of user preferences, allowing the system to determine likely purchases before they actually happen, thus expanding the data quantity beyond just sales records.
Solution Approach 2:
The patent implements continuous tracking of user interactions throughout the entire shopping session, from initial search to final purchase. This continuous data collection ensures that every user action contributes to the refinement query, maximizing the quantity of useful data captured while maintaining high reliability in preference determination through comprehensive coverage of the user journey.
3Productivity
If the system uses generic search terms to broaden search results, then the system can present more products, but the system cannot determine what the user is looking for
Solution Approach 1:
The system implements feedback mechanisms where user interactions with presented products (views, selections, additions to cart, purchases) are fed back into the query refinement process. This feedback loop allows the system to start with generic search terms to present many products initially, then progressively narrow down based on observed user behavior, maintaining both productivity and precision throughout the shopping session.
Data Source
AI summary
Techniques for determining search results based on session based refinements are presented herein. A method is disclosed that includes receiving a query in a user session, the query comprising one or more search parameters, detecting, in the user session and after receiving the query, a user event associated with a property of an item, updating a record in a table that associates the query with the property, the table comprising a plurality of records that associate the query with respective item properties, the record comprising the query, the property, and a score, and ranking search results for a subsequent query based on the associated properties indicated in the plurality of records, the subsequent query including the one or more search parameters.