Search Query Correction Using User Feedback for Uncommon Terms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing query correction systems often fail to provide relevant corrections for uncommon words or terms, leading to incorrect rankings and unsatisfied customer searches, resulting in lost sales and poor shopping experiences.
Innovation Solution
A system that analyzes user interactions with query corrections, updating databases to improve future corrections by storing pairs of search queries and their corrections or re-typed queries, and using machine learning models to enhance query correction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If rule-based methods (n-gram frequency) are used to determine query corrections, then common words can be corrected effectively, but uncommon words such as brands and television character names cannot be corrected accurately
Solution Approach 1:
The system uses feedback from user interactions with query corrections to continuously improve the ranking algorithm. User engagement data (acceptance, rejection, re-typing) is collected and used to re-rank candidate corrections, allowing the system to adapt to uncommon words and terms that were not present in the initial training data
Solution Approach 2:
The system changes the parameters used for ranking corrections from static n-gram frequency to dynamic models that incorporate user engagement data. This allows the ranking to adapt based on actual user behavior patterns, improving accuracy for uncommon terms while maintaining performance on common words
2Productivity
If query corrections are provided based on limited word corpus, then processing speed is maintained, but relevant corrections for customer intentions are missed
Solution Approach 1:
The system performs preliminary actions by collecting and storing user engagement data in databases (query-correction database and typo-candidate database) for future use. This preliminary data collection enables the system to provide more reliable corrections without requiring real-time analysis of all possible corrections
Solution Approach 2:
The system introduces intermediary databases that store historical query-correction pairs and user engagement data. These intermediaries act as mediators between the search query and the final correction decision, allowing the system to leverage historical patterns to improve reliability while maintaining processing speed
3Productivity
If wrong corrections are ranked at the top position, then processing efficiency is maintained, but customer satisfaction decreases and sales are lost
Solution Approach 1:
The system implements feedback mechanisms that monitor user interactions with corrections and use this information to improve future ranking decisions. By continuously learning from user acceptance, rejection, and re-typing behavior, the system aligns efficiency with improved customer experience
Solution Approach 2:
The system performs self-service by automatically learning from user interactions and improving its own correction ranking without external intervention. This self-improving mechanism allows the system to maintain efficiency while progressively improving customer satisfaction and reducing sales losses
Data Source
AI summary
This application relates to apparatus and methods for automatically determining query corrections based on prior interactions of users with a search query. In some examples, a computing device receives a search query from a user, and returns a query correction to the user in response to the search query. The computing device obtains engagement data corresponding to the query correction from the user. The computing device also updates one of a query-correction database and a typo-candidate database based at least in part on the engagement data. When the computing device receives the search query from another user, it corrects the search query from the another user using the query-correction database and the typo-candidate database.