Client-Side ML Ranking for Fast, Private Local-Remote Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search engines face challenges in determining relevant search results quickly while balancing privacy concerns and efficiently handling local and remote data, which complicates the ranking process.
Innovation Solution
Implementing client-side search ranking improvements using machine learning models to filter, score, and rank local and remote search results based on computed features, applying configurable decision rules and transforming data to enhance relevance and privacy preservation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple fields, features and parameters are used to determine relevant search results, then the relevance of search results is improved, but the complexity of the ranking process increases
Solution Approach 1:
The patent segments the ranking process into multiple independent machine learning models, each responsible for specific aspects of search result evaluation. These models process different feature sets independently and their outputs are combined to produce the final ranking, thereby managing complexity while maintaining comprehensive relevance assessment
Solution Approach 2:
The patent introduces machine learning models as intermediary components between raw search data and final ranking results. These models act as mediators that automatically process and interpret multiple fields, features and parameters, reducing the manual complexity of the ranking process while improving relevance through data-driven insights
2Measurement precision
If comprehensive search analysis is performed to improve relevance, then the quality of search results is improved, but the response time increases
Solution Approach 1:
The patent performs preliminary processing of search data by pre-computing features and training machine learning models in advance. This allows the system to quickly apply pre-trained models to new search queries without performing comprehensive analysis from scratch, thereby maintaining high quality results while reducing response time
Solution Approach 2:
The patent implements a two-stage ranking approach where a first machine learning model performs a quicker partial ranking to identify candidate results, and a second model performs more comprehensive analysis on a smaller subset. This partial action strategy maintains quality while significantly reducing overall response time
3Measurement precision
If user information is collected for personalized search ranking, then the relevance of search results is improved, but privacy concerns increase
Solution Approach 1:
The patent implements local machine learning models on user devices that process and analyze user information locally rather than transmitting it to remote servers. This allows personalized search ranking based on user-specific data while maintaining privacy by keeping sensitive information on the local device where it can be used for ranking without external exposure
4Quantity of substance
If local and remote data are integrated for search, then the completeness of search results is improved, but the complexity of data handling increases
Solution Approach 1:
The patent merges local and remote search data processing into a unified framework where machine learning models handle both data sources through consistent feature computation and ranking logic. This integration approach maintains completeness by combining multiple data sources while managing complexity through standardized processing pipelines
Data Source
AI summary
Methods and systems for client side search ranking improvements are disclosed. In one example, a search query is received from a user on a client device. The local search results are filtered based on the received search query and one or more local ranking rules. Features for each filtered local search result are computed. The computed features of each filtered local search result are input to one or more machine learning (ML) models. Each ML model can generate a score for each filtered local search result. The filtered local search results are ranked based on the generated score within a category. In one example, local search results and remote server search results are obtained. The local search results and remote server search results are ranked using at least one machine learning (ML) ranking model. The ranked local search results and remote server search results are displayed on the client device by category.